Phage display has been used extensively in vitro and in animal models to generate ligands and to identify cancer-relevant targets. We report here the use of phage-display libraries in cancer patients to identify tumor-targeting ligands. Eight patients with stage IV cancer, including breast, melanoma, and pancreas, had phage-displayed random peptide or scFv library (1.6 × 108-1 × 1011 transducing units/kg) administered i.v.; tumors were excised after 30 minutes; and tumor-homing phage were recovered. In three patients, repeat panning was possible using phage recovered and amplified from that same patient's tumor. No serious side effects, including allergic reactions, were observed with up to three infusions. Patients developed antiphage antibodies that reached a submaximal level within the 10-day protocol window for serial phage administration. Tumor phage were recoverable from all the patients. Using a filter-based ELISA, several clones from a subset of the patients were identified that bound to a tumor from the same patient in which clones were recovered. The clone-binding to tumor was confirmed by immunostaining, bioassay, and real-time PCR–based methods. Binding studies with noncancer and cancer cell lines of the same histology showed specificity of the tumor-binding clones. Analysis of insert sequences of tumor-homing peptide clones showed several motifs, indicating nonrandom accumulation of clones in human tumors. This is the first reported series of cancer patients to receive phage library for serial panning of tumor targeting ligands. The lack of toxicity and the ability to recover clones with favorable characteristics are a first step for further research with this technology in cancer patients. (Cancer Res 2006; 66(15): 7724-33)
- phage display
- cancer patients
- ligand selection
- tumor binding
- targeted therapy
Approval of trastuzumab (Herceptin; Genentech, South San Francisco, CA) in September 1998 by the Food and Drug Administration (FDA) for treatment of breast cancer was a major clinical milestone in the field of targeted therapeutics. Binding of ErbB2 with Herceptin leads to inhibition of proliferation of cancer cells as shown in preclinical ( 1, 2) and clinical studies in the metastatic and adjuvant setting ( 3– 5). The Herceptin molecule is modular in the sense that the target binding portion forms one module and the effector portion another. Extensive molecular modification of the effector module has been done to render Herceptin an effective anticancer drug.
Non-human-derived antibodies are a rich source of tumor-binding reagents. As a therapeutic reagent, the tumor-binding portion of nonhuman antibodies is the molecular feature of interest. The non-tumor-binding portion is replaced or linked with a molecular module that offers improved features, which may include decreased immunogenicity, improved pharmacokinetics, or increased antiproliferation capacity. An alternative source of tumor-binding reagents, similar to the binding portion of the antibody molecule, is from phage display technology ( 6). Although a variety of formats have been developed, the most common phage display system expresses the fusion protein on the minor coat protein, pIII ( 7), or the major coat protein, pVIII ( 8, 9). The size of the fusion protein varies from a short peptide to antibody fragments. Phage display technology has been highly successful in generating ligands against a large variety of targets, including cell receptors ( 10), tumor-associated antigens ( 11), hormones and cellular messengers ( 12), matrix-related elements ( 13), viruses ( 14), immunoglobulins ( 15), DNA ( 16), autoimmune antigens ( 17), drugs ( 18), intracellular signal transduction molecules ( 19), and clotting factors ( 20). Although the initial binding unit discovered through phage display may have agonist or antagonist biological activity, frequently a bioactive module must be added to the phage product to obtain the desired effect.
The majority of research with phage-displayed ligand libraries (PDL) has involved screening against purified targets bound to a solid surface. A notable exception has been the use of phage display technology for screening in vivo with animals ( 21– 25). After i.v. injection of PDLs, animals were sacrificed, target organs harvested, and the collected phage were administered to another mouse to enrich phage that bound to the desired target. These in vivo panning strategies have resulted in selective ligands to the vasculature of various organs ( 23, 24, 26, 27), tumor xenograft ( 21, 28), and tumor lymphatics ( 29). A recent report showed strong tumor selectivity after only two in vivo pans in a mouse transgenic model of thyroid carcinoma ( 30). In addition, phage peptides were identified that interact with blood vessel targets of a particular organ following single i.v. infusion of phage-displayed random peptide library (RPL) to a brain-dead patient ( 31). These novel experiments represent an entirely new approach to identification of tumor-specific ligands and the cognate targets.
Our lab has focused on in vivo panning strategies in which the panned organism is not sacrificed or allowed to die after the screening event. Our experimental studies in mice provided the necessary preclinical information for conducting serial PDL infusions ( 32). We now report here a first set of cancer patients that have participated in a phase I study and have received PDL infusions. This is the first report describing living cancer patients receiving serial infusions of phage libraries with recovery and reinfusion of tumor-bound phage. A successful identification of tumor-specific ligands in the present study opens a new field of tumor ligand discovery that has direct application to the clinical setting.
Materials and Methods
This clinical study was done in accordance with the FDA and the University of Vermont Institutional Review Board Committee on Human Research in the Medical Sciences.
Eligibility criteria include stage IV malignancy of any histology, cancer nodules, or mass amenable to biopsy or resection with relatively minor surgery; ≥18 years of age; Karnofsky status ≥70; life expectancy ≥4 months; not pregnant; serum laboratory values related to hematology (hemoglobin ≥10 g%, hematocrit ≥30%, absolute neutrophil count ≥1,500/μL, platelets ≥100,000/μL); renal creatinine less than upper limit of normal and hepatic function less than twice the institutional upper limit of normal; cardiac status New York Heart Association grade 2 or less; pulmonary function at least by forced expiratory volume and/or diffusing capacity for carbon monoxide 60%; no clinical symptoms of brain metastasis; no concurrent cytotoxic drugs or radiation therapy during the week preceding or the week after phage infusion; and nonelevated anti–filamentous phage antibody titer.
Two phage-displayed RPLs and one single-chain antibody (scFv, Tomlinson I) library were used for human administration. The peptide/scFv sequences in these libraries were expressed as NH2 terminus fusion to minor coat pIII protein of filamentous bacteriophage. The phage-displayed RPL used in these studies was constructed in the fUSE5 gene III phage-display system ( 15, 33). The fUSE5 vector and E. coli host strains were a generous gift from Dr. George Smith (University of Missouri, St. Louis, MO). The half-site cloning method used by Cwirla et al. ( 15) was employed in RPL construction. One peptide library displayed 12 random amino acids with two cysteine residues fixed at positions 1 and 14 attached to a linker (CX12CG4SG3A2, where X is any amino acid) and the other 20-mer with cysteines fixed at positions 5 and 16 (X4CX10CX4). The complexities of the random 12- and 20-mer libraries were 1.26 × 107 and 1.8 × 107 different peptides, respectively. The details of library preparation, titration, elution of cell-bound phage, phage plating, and amplification have been described in detail in our earlier publications ( 19, 32, 34, 35).
The human single-fold scFv Tomlinson I library ( 36), cloned in ampicillin-resistant phagemid vector pIT2 and transformed into TG1 E. coli cells, was obtained from MRC, HGMP Resource Centre (Hinxton, Cambridge, United Kingdom). The scFv fragments compose a single polypeptide with the VH and VL domains attached to each other by flexible glycine-serine linker. scFv library has diversity of 1.47 × 108. The scFv library amplification, titration, elution of cell-bound phage, and their plating were done as previously described ( 36, 37) using KM13 helper phage ( 38) and TG1 E. coli bacteria.
Preparation of Phage Library for Infusion
The PDLs were prepared and tested according to FDA standards. Endotoxins were removed from the preparation by 1% (v/v) Triton X-114 extractions until the preparation had endotoxin levels under the FDA limit of 5 endotoxin units/kg as determined by Limulus Amebocyte Lysate gel clot assay (Endosafe, Charles River Laboratories, Wilmington, MA). The phage were concentrated with polyethylene glycol and completely solubilized in PBS containing protease inhibitor aprotonin (Trasylol, Bayer Pharmaceuticals Corporation, West Haven, CT). Phage was passed through a pyrogen-free 0.2-μm polyethersulfone membrane filter. Sterility of the phage preparation that was injected into the human patient was tested according to the U.S. FDA Code of Federal Regulations (21 CFR 610.12) using fluid thioglycollate and soybean casein media.
Estimation of Serum Phage Antibodies
Serum was collected for evaluation of phage antibodies before each infusion and at 4 and 8 weeks after the last phage infusion. This was carried out using an ELISA-based method developed and used in our preclinical study ( 32). Briefly, a 96-well polystyrene microplate (F96 Maxisorb Immuno plate, Nulge Nunc, Roskilde, Denmark) was coated with 1 × 107 transducing units (TU) of filamentous phage (from phage library) overnight at 4°C. The unbound phage were removed with 0.1% Tween 20-TBS buffer and blocked with 1% casein in TBS. PBS, human immunoglobulin G (IgG; negative control), and the serum samples (1:1,000) were added to the appropriate wells (100 μL/well). These samples were incubated on the plate for 2 hours at room temperature. The plate was washed again to remove any unbound material. Goat anti-human IgG-horseradish peroxidase (HRP) conjugate (1:10,000) was added to the wells and incubated for 2 hours at room temperature. The plate was washed again to remove the unbound IgG. The ABTS substrate was used for detection of the HRP conjugate.
Infusion and Tumor Biopsy
Infusions and surgical excisions were done with the support of the University of Vermont General Clinical Research Center in a room dedicated for this protocol with appropriate surgical equipment, sterile supplies, monitoring equipment, and nursing personnel. The phage aliquot was added to 100 mL of physiologic saline and i.v. administered over 15 minutes. We aimed to slowly increase the phage dose with each new infusion after experiencing the safety of previous dose. However, the yield of final endotoxin-free phage preparation, particularly with fUSE5 expressing peptide library, was a limiting factor in following predetermined increases in the phage doses. Thirty minutes following infusion, a superficial tumor nodule was surgically removed. Vital signs, including blood pressure, pulse, respiration, pulse oximetry, and level of consciousness, were monitored closely throughout the infusion, biopsy, and for 2 hours after the procedure. Patients were reevaluated at 1 and 2 months following the procedure.
Tumor Processing and Phage Recovery
The tumor was immediately processed following excision. The tumor was rinsed to remove blood and cut into several small pieces. It was further rinsed five times by repeating suspension and centrifugation steps. The rinsed tissue and cells were homogenized to release cell-internalized phage. The bacterial host infection, plating, titration, and amplification were done by the methods as described earlier for peptide ( 19, 32, 34, 35) and scFv ( 36, 37) libraries. Phage administration, tumor harvest, and amplification were repeated twice for a maximum of three infusions. All infusions were completed within 10 days to avoid significant patient antibody response to phage ( 39).
Binding Assays of Phage Clones Recovered from Patient Tumor
The tumor-homing phage clones from two patients, one each following peptide (180-05) and scFv library (180-08) infusion, were assessed for their binding to the tumor cells from the same patient in which the clone was recovered. The amount of preinfusion tumor tissue was a limiting factor in determining the number of tumor-homing clones assessed for the binding. For binding assessment, individual clones directly recovered from a tumor were randomly selected and amplified for phage production. To use equal amounts of phage particles for binding assessment of different clones, the relative concentrations of phage in supernatants were determined by a chemiluminescence ELISA method as previously described ( 40).
Chemiluminescence ELISA for binding screening. Initial evaluation of tumor-homing clones for binding to the same patient's tumor cells was done with an in vitro assay that we have optimized for this purpose ( 41). A key attribute of this assay is that it does not require cells to be bound to a surface of a well or a bead. Microtiter 96-well format plates with a filter bottom allow retention of cells but also rinsing and removal of unbound phage. In this assay, cells are introduced to the filter-bottom wells and exposed to each selected phage clone. Following incubation and appropriate rinses, binding is detected with an HRP/anti-M13 monoclonal conjugate and a chemiluminescence HRP substrate. The clones that showed considerable binding to patient's own tumor cells were further studied for their binding to patient's own blood cells and to the tumor cells from other patients. In addition, these clones were also studied for their binding to certain human nontumor (melanocytes and Hs 578Bst) and cancer (breast: SK-BR-3, T47D, BT474, MDA-MB-231, and MDA-MB-361; melanoma: LOXIMVI, M14, SK-MEL-2, SK-MEL-28, UACC-62, and UACC-257) cell lines of the same histology. Breast and melanoma cell lines were obtained from American Type Culture Collection (Manassas, VA) and National Cancer Institute, respectively.
Immunostaining evaluation of binding. Following antigen retrieval in citrate buffer (pH 6) at 90°C, tumor cells were washed and blocked overnight with 1% casein at 4°C. The next day, ∼10,000 tumor cells were transferred to filter cups of a 24-well Costar transwell plate (Corning, Inc., Corning, NY) having 6.5-mm diameter cups with polycarbonate membrane (8 μm) bottom. The cells were incubated with unselected library phage or a tumor-homing clone in incubation buffer (TBS, 0.05% Igepal, 0.1% bovine serum albumin) for 2 hours with gentle shaking. Cells were washed eight times with TBS containing 0.05% Igepal (TBS-Igepal) and incubated with mouse anti-M13 (Amersham Pharmacia Biotech, Little Chalfont, United Kingdom), diluted 100 times in incubation buffer, for 1 hour. After washing five times with TBS-Igepal, the cells were incubated for 1 hour with goat anti-mouse/Alexa Fluor 488 or 568 conjugate (Molecular Probes, Carlsbad, CA) in incubation buffer (1:100). Cells were washed six times with TBS-Igepal before fluorescence microscopy.
Binding assessment using biological titration and real-time PCR. One hundred thousand tumor cells following antigen retrieval step were dispensed to casein-blocked 1.5-mL tubes and incubated with different phage clones normalized for their concentrations. Tubes were shaken slowly at room temperature for 2 hours. Cells were washed four times with TBS-Igepal and once with TBS to remove unbound phage. The washed cells were mixed with E. coli for biological titration as described earlier. The quantification of cell-bound phage was also done by real-time PCR analysis of heat-released phage DNA using appropriate primers and probes ( 42).
Insert DNA sequence analysis. The double-stranded DNA of tumor-homing phage clones was isolated using the QIAprep Spin Miniprep kit (Qiagen, Inc., Valencia, CA). The sequencing reactions and automated DNA sequencing were done in the Vermont Cancer Center DNA Analysis Facility using BigDye Ver.1 Dye Terminator kit and appropriate primers. Amino acid sequences of insert peptides were deduced from the DNA sequence of the corresponding phage clones. Consensus sequence identification was done using Pileup, Seqlab, and Prettybox tools of the GCG Wisconsin Package (Madison, WI). National Center for Biotechnology Information and European Molecular Biology Laboratory-European Bioinformatics Institute databases were searched for biological protein sequences that match with identified ligands. All sequences were analyzed according to their sequence similarity from a single patient tumor.
Eight patients have participated in this phase I clinical study of i.v. phage display screening for tumor-binding ligands ( Table 1 ). The number of phage infusions varied from one to three. The tumor types included breast cancer, malignant melanoma, and pancreas cancer. The ages ranged from 43 to 72 years. All tumors were in a relatively superficial location. The amount of phage administered i.v. ranged from 1.6 × 108 to 1.0 × 1011 TU/kg per infusion. The number of tumor nodules removed varied from one to four. The locations of the tumor nodules excised are described in Table 1.
Immediate side effects during infusion were observed in only one patient (180-05) during the second infusion. The side effect was grade 1 to 2 pain in the upper mid-back after 75% of the phage had been infused. The infusion was immediately stopped and the pain fully subsided after 1 to 3 minutes. No other side effects attributable to the infusion were observed. All observed events were evaluated by an independent medical oncologist data safety officer and were determined to be due to progressive cancer. All patients were stage IV and the majority had exhausted available systemic therapy options. The most significant concern was immediate allergic reactions and none were observed.
Before phage infusion, none of the eight participating patients in the study had elevated antiphage IgG levels. Serum IgG levels against phage increased in all patients after phage infusions over the maximum monitoring period of 75 days. The data from six patients are presented in Fig. 1 . Two patients could not be followed for their postinfusion serum IgG analysis owing to urgent attention to cancer-related complications (180-09) and patient (180-07) moving out from the study. The follow-up serum antiphage IgG measurements could not be done at the same time intervals in different patients because of the unavailability of patients for blood collection. The data show that a full immune response against filamentous phage did not develop within the 10-day interval to which the serial phage infusions were restricted.
Phages were recovered from every patient that underwent infusion and recovery was increased with increasing doses. Five patients had a single infusion, two patients had two, and one patient had three infusions. The number of infusions varied according to the number of tumor nodules amenable to biopsy. The clone analysis from the patients with more than one infusion for their insert peptide sequences revealed the recovery of a less diverse clone population in subsequent rounds of infusion. Furthermore, we observed motif sharing among insert amino acid sequences of tumor-homing clones and identification of those motifs in human proteins. As an example, we present here the sequence data of one patient. Table 2 shows amino acid sequences, their frequency, and possible motifs of phage clones obtained from surgically harvested breast cancer tumor from patient 180-04 following second infusion. An enrichment of tumor-specific clones is indicated by the multiple occurrences of certain clones. A search of Swiss protein database for human proteins matching the motif sequences from this patient (180-04) revealed several significant matches ( Table 3 ).
The binding assessment of clones recovered from the tumors of two patients, one each following peptide (180-05) and scFv (180-08) library infusion, was done on tumor cells from the same patient in which panning was done. The number of clones assessed per patient was limited according to availability of patient tumor cells and the number of clones recovered. Figure 2A shows the results of the binding assay of randomly selected tumor-homing peptide clones recovered from the first round of panning in patient 180-05 with malignant melanoma. These results show a positive binding by clone 1922 to the tumor cells from the same patient in which the clone was recovered. We considered a binding positive if it showed thrice or more binding than that of unselected library clones. The results of binding screening of randomly selected tumor-homing clones from patient 180-08 following scFv library infusion are presented in Fig. 2B. The data clearly show positive binding by several scFv clones to breast tumor cells from the same patient in which the clone was recovered.
The binding of positive peptide clone 1922 and six other scFv clones (701, 718, 729, 730, 761, and 775), which showed considerable binding in chemiluminescence ELISA, was confirmed by using other procedures as well. The immunostaining of selected peptide clone 04-1922, as presented in Fig. 3A , clearly shows higher binding to patient's melanoma cells, as compared with that shown by unselected library clones. Figure 3B presents the immunostaining of scFv clones showing significantly higher binding to tumor cells from the same patient in which the clone was recovered, as compared with unselected library clones or an irrelevant clone. The positive binding of clones was also confirmed by quantifying cell-bound phage using real-time PCR and bioassay methods (data not presented).
The tumor-binding peptide clone 1922 from patient 180-05 and six scFv clones (701, 718, 729, 730, 761, and 775) from patient 180-08 were also studied for their binding to patient's own blood cells, other cancer patients' tumor cells, and different tumor and nontumor cells lines for determining cell specificity. These tumor-binding clones did not show positive binding to the blood cells of the patient from which they were isolated. The binding studies of clone 1922 on patient 180-08 tumor cells and that of the six scFv clones on patient 180-05 tumor cells showed that these clones did not bind to the tumor cells from which they were not isolated. Furthermore, five of the six breast tumor-positive scFv clones did not show any binding to human breast nontumor (Hs 578Bst) and five tumor cell lines (SK-BR-3, T47D, BT474, MDA-MB-231, and MDA-MB-361). Clone 718 showed some binding to MDA-MB-361, MDA-MB-231, and BT474 breast cancer cells that, depending on the cell line, were 23% to 56% of the binding observed in patient's breast tumor cells. The peptide clone 1922 from melanoma patient 180-05 showed little binding to SK-MEL-5 melanoma cells but did not show any binding to human melanocytes and six other NCI-60 melanoma cell lines tested (LOXIMVI, M14, SK-MEL-2, SK-MEL-28, UACC-62, and UACC257; data not presented).
DNA of tumor-binding peptide clone 1922 and the six scFv clones were sequenced to determine their amino acid sequences. DNA analysis showed that tumor-binding scFv clones 701, 718, 729, 730, 761, and 775 were unique. Swiss protein database search of the insert peptide sequence (MRIRCAAAWRATGTHCSLRA) of clone 1922 revealed that this peptide shares a significant motif with human multiple epidermal growth factor–like domain protein 7 (accession no. Q9UHF1).
The highlighted motif-sharing regions having amino acids with exact matches are shown in white letters and those with conservative substitutions are in black letters.
Our long-term goal is to develop a process for generating a set of tumor-binding ligands for an individual patient. The initial strategy used here was to administer a phage library i.v., surgically sample tumor tissue, and collect phage that have bound to the tumor. This work follows preclinical toxicity testing ( 32) and approval by the FDA to conduct a phase I study in human cancer patients. As a phase I study, our focus in the first three patients was on the safety of phage infusion only. In the fourth patient (180-04), we obtained a large number of clones which were sequenced and analyzed for possible motif sharing among them. Later in other two patients with sufficient preinfusion tumor tissue, we conducted clone binding studies. We present here our preliminary clinical data showing (i) safety of phage display library administration; (ii) recovery of phage clones from patient tumors; (iii) identification of tumor-binding clones; (iv) cell-specificity of tumor-binding clones; and (v) consensus sequences of phage clones consistent with human proteins. The benefits of this procedure in human cancer patients include the presence of the complete set of possible targets which are otherwise not available in vitro, targets in their native configuration, subtraction of ligands that bind to normal tissue, identification of ligands that will be stable in blood, the possibility of identifying unique targets for an individual patient, and a rapid process.
The application of ligands derived from human panning is an important component of a modularized drug concept. One module is the tumor-binding element derived from the panning process and the other is the antiproliferative component. A wide variety of options are available for the antiproliferative module, which include virtually every current anticancer drug in which increased accumulation of drug at the cancer site would be advantageous. Examples of possible antiproliferative modules include cytotoxic, gene therapy delivery, immunologic, radioactive, apoptotic, antimetabolic, and interruption of critical signal transduction pathways. Whichever strategy is subsequently chosen will be of considerable benefit if tumor-binding ligands for any patient are reliably available. Our strategy is to first focus on developing methods to produce customized ligands for cancer patients without immediate regard to the choice of an antiproliferative component.
Human cancer patient panning is necessarily constrained by the need to not harm the individual either through toxicity or through planned procedures that would be relatively morbid. Panning in animal models has much less constraint and several studies have been done in which the animal is sacrificed with each panning step. In vivo panning in animals using this strategy has led to identification of several organ-specific ligands and, in some cases, their important cognate targets ( 21, 23). Blood vessels of normal organs are the most likely targets in an experimental panning scheme in which a brief time is allowed between phage administration and organ harvest because penetration through the blood vessels would be relatively low. However, tumors have a vasculature that is much different than that of normal organs. In addition to unique blood vessels targets, phage would have opportunities for direct contact with cancer cells, interstitial cells, and matrix elements ( 43). Using transgenic and xenograft animal models, ligands have been identified to both tumor vasculature ( 21) as well as tumor cells ( 28). We have focused on adapting the same strategy to human cancer patients.
A total of 12 phage infusions were done in eight patients. No clinically detectable allergic reaction was identified. One patient had mild to moderate discomfort during a second infusion in the upper back region that lasted for <5 minutes. The exact etiology of this discomfort was unclear but was related to the infusion because it occurred during an infusion and resolved immediately on cessation of the infusion. Consistent with our preclinical data, no other significant adverse events could be identified related to the phage infusions. Although not in a displayed library format, delivery of ϕ-X 174 bacteriophage in vivo has been accomplished in >3,000 individuals as a reagent for evaluation of immune function ( 39, 44– 47). Serious adverse reactions were not reported among these patients. Although this large experience has been with a different type of bacteriophage than that used in our study, it indicates a reasonable safety profile of administering bacteriophage to humans. In addition, all patients in our study had no detectable antiphage IgG levels before infusion. The follow-up studies with postinfusion serum IgG levels indicated that a significant immune response did not develop within 1 week, to which the serial phage infusions should ideally be restricted. Serum IgG levels against filamentous phage increased over 3 to 8 weeks following phage infusion.
Assessment of the ability of clones to bind to the type of tissue the clones were harvested from is a critical step in identifying tumor-binding clones. In animal models, this is a relatively simple task because binding assays can be done in vitro with cell lines from the tumor. Alternatively, binding assays can be done in vivo by injection into additional animals. Neither of these options is readily available after human cancer patient panning. The binding assays with cell lines are generally done with cells attached to a well allowing rinsing and removal of nonbinding phage. Cancer cells recovered from patients will not reliably adhere to a plastic surface, negating the option of this in vitro approach. We have applied a separation scheme in which the patient tumor cells are retained on a filter and the nonbinding phage are removed through rinsing ( 41). Phage bound on the tumor cells are quantitated with antiphage antibodies. This procedure is amenable to a microplate assay, facilitating miniaturization of the procedure. In addition, phage clone binding can be assessed with very small amounts of tumor tissue. Given the finite amount of patient tumor available, this approach is valuable in assessing larger numbers of candidate clones. Figure 2 shows the clone binding assessment on the tumors from two patients using this procedure. A number of clones showed considerable binding in comparison with naive library clones as negative control. It is interesting to note a higher ratio of hit clones from scFv-infused than peptide-infused patient. This may be related to a higher dose of scFv library (1 × 1011 TU/kg) than RPL (1.6 × 109 TU/kg) infusion. It is also possible that certain structural characteristics of scFv library work in favor of better target binding. In the past also, we had a better hit ratio with scFv library than RPL in in vitro panning with cancer cells ( 41). The positive binding of these clones to tumor cells from the same patient in which the clone was recovered was further validated by other assay methods. These tumor-binding clones did not bind to the patient's own blood cells, ruling out the possibility of selected clones being specific to HLA type of that patient. Furthermore, the observations of selected clones not binding to the tumor cells of other patients from which they were not isolated showed some degree of cell specificity. The binding specificity of these clones was further assessed by using several human noncancer and cancer cell lines. Of six selected breast tumor-homing scFv clones, only clone 718 showed significant binding, though lower than patient tumor cells, to MBA-MD-231, BT474, and MDA-MB-361 breast cancer cells. Others did not show significant binding to any of the breast cell lines tested. Similarly, clone 1922 from melanoma patient did not bind to human melanocytes and several NCI-60 melanoma cell lines, except for SK-MEL-5, which showed little binding. It seems that the target(s) on patient tumor cells, which binds to clone 718 or 1922, is overexpressed in some established cancer cell lines as well.
An alternative approach to identification of ligands and for identification of potential targets is to search for consensus sequences among the recovered phage clones. Redundant clones will be identified, minimizing unnecessary downstream work on identical clones, and often clones present in the highest frequency are better binding clones. Sequencing becomes more valuable in situations where binding assays are more challenging to do. Whereas sequencing information provides no direct information on binding, strong sets of consensus sequences are suggestive of binding events to distinct targets. Indeed, this approach has been reported in a person that met the criteria for brain-based determination of death ( 31). In our human cancer patient series, we sequenced hundreds of clones recovered from tumors following phage infusion and analyzed them for motif sharing. We presented amino acid sequences of the clones recovered from metastatic breast cancer nodule in patient 180-04 as an example of such studies. The presence of multiple sequences of several clones clearly indicates an enrichment of clones following second round of panning. Clone 269 seems to be very interesting as its prevalence among the clones sequenced from this screen is very high. Furthermore, in addition to the two cysteines from the library design, this clone contains two new cysteines, giving it an unusual dicyclic structure. However, binding studies with this and other clones from this screen could not be done due to limited availability of preinfusion tumor tissue that was consumed in initial trouble shooting experiments and screening few clones from the first round of infusion. Several strong motifs were identified, indicating that the presence of clones in the breast cancer nodule was nonrandom. These motifs, as was also true of peptide clone 1922, shared significant sequence similarity with a variety of human proteins. The proteins with shared sequences are very interesting in that several were related to overexpression in cancer cells or were related to cancer phenotypes. EGFL7, which shares motif with melanoma cell-binding clone 1922, is expressed at high levels in the vasculature associated with tissue proliferation. Loss of its function in EGFL7 knockdown zebra fish embryos has been reported to specifically block vascular tubulogenesis ( 48). Presently, it is difficult to ascertain the importance of these motifs in recognizing phenotypic and/or physiologic characteristics of the patient's tumor.
In conclusion, this is the first phase I clinical report of bacteriophage library infusions in human cancer patients. No significant toxicity was observed in this preliminary set of patients. Patients had no detectable antiphage serum IgG before infusion and did not develop maximum antibody response over the 10-day study infusion limit. Several amino acid motifs were identified among tumor-homing phage clones, indicating nonrandom tumor accumulation. Phage clones were identified that bound to tumor from the patient in which the infusion was administered. This phase I study indicates the feasibility of this technology for possible identification of customized patient tumor-binding ligands.
Grant support: National Cancer Institute grant R21 CA97679 (D. Krag); Department of Defense grant DAMD 17-99-1-9425 (D. Krag; views and opinions of and endorsements by the author(s) do not reflect those of the U.S. Army or the Department of Defense); General Clinical Research Center grant MO1 RR00109; Vermont Cancer Center Support grant PHS P-30 22435; and SD Ireland Cancer Research Foundation.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Received December 13, 2005.
- Revision received May 5, 2006.
- Accepted June 6, 2006.
- ©2006 American Association for Cancer Research.