Identification of Clinically Significant Tumor Antigens by Selecting Phage Antibody Library on Tumor Cells in Situ Using Laser Capture Microdissection*S

Much work has been done to develop tumor-targeting antibodies by selecting a phage antibody library on cancer cell lines. However, when tumor cells are removed from their natural environment, they may undergo genetic and epigenetic changes yielding different surface antigens than those seen in actual cases of cancer. We developed a strategy that allows selection of phage antibodies against tumor cells in situ on both fresh frozen and paraffin-embedded tissues using laser capture microdissection. By restricting antibody selection to binders of internalizing epitopes, we generated a panel of phage antibodies that target clinically represented prostate cancer antigens. We identified ALCAM/MEMD/CD166, a newly discovered prostate cancer marker, as the target for one of the selected antibodies, demonstrating the effectiveness of our approach. We further conjugated two single chain Fv fragments to liposomes and demonstrated that these nanotargeting devices were efficiently delivered to the interior of prostate cancer cells. The ability to deliver payload intracellularly and to recognize tumor cells in situ makes these antibodies attractive candidates for the development of targeted cancer therapeutics.

Much work has been done to develop tumor-targeting antibodies by selecting a phage antibody library on cancer cell lines. However, when tumor cells are removed from their natural environment, they may undergo genetic and epigenetic changes yielding different surface antigens than those seen in actual cases of cancer. We developed a strategy that allows selection of phage antibodies against tumor cells in situ on both fresh frozen and paraffin-embedded tissues using laser capture microdissection. By restricting antibody selection to binders of internalizing epitopes, we generated a panel of phage antibodies that target clinically represented prostate cancer antigens. We identified AL-CAM/MEMD/CD166, a newly discovered prostate cancer marker, as the target for one of the selected antibodies, demonstrating the effectiveness of our approach. We further conjugated two single chain Fv fragments to liposomes and demonstrated that these nanotargeting devices were efficiently delivered to the interior of prostate cancer cells. The ability to deliver payload intracellularly and to recognize tumor cells in situ makes these antibodies attractive candidates for the development of targeted cancer therapeutics.

Molecular & Cellular Proteomics 5:2364 -2373, 2006.
Due to ease of accessibility, tumor cell surface antigens are invaluable targets for therapeutic development. The epitope space at the cell surface is highly complex. Relevant antigens may include glycosylated proteins and other post-translationally modified products that may not be readily predicted from studies of genomic copy number or mRNA expression levels (1)(2)(3)(4)(5)(6). Because monoclonal antibodies (mAbs) 1 recognize a wide range of antigenic determinants with high affinity and specificity and are able to discern subtle differences in antigen structure and conformation, they can be used to efficiently map the tumor cell surface epitope space (1). Isolating these epitopes enables the antibodies to achieve specific binding to neoplastic cells, an ability that could be utilized in applications such as induction of antibody-dependent cell cytotoxicity (7) or inhibition of signaling pathways involved in tumor cell migration, growth, and survival (8,9). In addition, antibodies targeting internalizing tumor epitopes could be exploited to achieve efficient and specific intracellular delivery of chemotherapeutic drugs or other tumor-modulating agents (1, 10 -12).
Phage antibody display has been widely used to develop cancer-specific antibodies (1,(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23). A combinatorial phage antibody library serves as a source of random shape repertoire that can be used to probe neoplastic variations on the surface of cancer cells (1, 24 -26). Selecting phage antibody libraries directly on cancer cell lines enables the identification of tumor-targeting antibodies without prior knowledge of target antigens (1,20,24,25). Although numerous antibodies have been found by this approach, the screening process against cell lines does not provide an ideal picture as to how specific these antibodies will be to actual cancer cells in patient populations. After several generations in culture, cancer cell lines may express cell surface epitopes that differ from those present in the original cancerous tissue. Tissue sections from cancer patients would be an ideal selection target in the development of cancer-specific antibodies; however, most tissues taken during surgeries, biopsies, or autopsies are composed of heterogeneous cell populations. This seemingly poses a serious obstacle to selection methods that would specifically target cancer cells in tissue.
The advent of laser capture microdissection (LCM) has allowed small clusters of homogenous cells to be isolated and removed from tissue sections under direct microscopic visualization (27)(28)(29)(30)(31). This technology therefore permits the selection of phage binding specifically to tumor cells while ignoring adulterating entities such as non-neoplastic cells and stromal elements. Using this technique in conjunction with phage antibody display, it is possible to select an antibody library against actual cases, generating a set of cancer-specific antibodies that are more likely to target clinically relevant epitopes.
We hereby describe the development of an LCM-aided selection strategy that allows the identification of phage antibodies that bind to internalizing cell surface antigens present on actual cases of human cancer. The ability to deliver payload intracellularly and to recognize tumor cells in situ makes these antibodies attractive candidates for the development of targeted cancer therapeutics.

Creating a Sublibrary Enriched for Binders to Functional Tumor Cell
Surface Epitopes-The sublibrary was created by selecting a naïve phage antibody display library on a panel of tumor cell lines under internalizing conditions. The preparation and selection of a phage antibody display library has been described previously (1,25). Briefly the phage library was preincubated with a panel of non-tumorigenic cells including BPH-1, human mammary epithelial cells, MCF10A, and human fibroblasts to remove binders to common cell surface antigens. The predepleted library was then incubated with a panel of prostate cancer cell lines (PC3 and Du-145) at 37°C for 2 h; washed twice with 100 mM glycine, pH 2.8, in the presence of 150 mM NaCl; and washed once with PBS, pH 7.0. Internalized phage were recovered by lysing the cells with 100 mM triethylamine, propagated in TG1, and purified by precipitation with polyethylene glycol 8000 as described previously (1), thereby creating a sublibrary that is enriched for binders to internalizing cell surface molecules. The sublibrary contained 1-5 ϫ 10 5 copies of about 10 6 independent clones at the concentration of 1-5 ϫ 10 11 cfu/ml.
Selection of Antibodies Targeting Tumor Cells in Situ by LCM-Selections were performed on both frozen and paraffin-embedded prostate cancer tissues. For selection on frozen tissue slides, 5-m sections from prostate cancer specimens were cut onto Leica Mem-braneSlides (MicroDissect, Mittenaar, Germany), stained with hematoxylin, and incubated with the sublibrary (0.5 ml of 5 ϫ 10 11 cfu/ml stock) at room temperature for 1 h. The slides were then washed three times in PBS to remove unbound phage and prepared for LCM by dehydration in 70, 95, and 100% ethanol in series. For selection on paraffin-embedded tissue, 5-m sections were cut onto film-coated Leica slides, xylene-treated to remove paraffin, rehydrated through serial 100, 95, and 75% ethanol, placed in PBS with blocking solution at room temperature for 1 h, washed, and incubated with the sublibrary described above. LCM was performed using the Leica AS LMD (Leica Microsystems GmbH, Wetzlar, Germany) that uses a UV pulse laser to excise selected cells from surrounding tissues. Typically 20 -50 tumor cells were procured at a time by generating a closed laser path around the group of cells of interest. The cells were then dropped into collection tubes by electrostatic force and gravity. These tissue pieces were stored at Ϫ80°C until analysis.
Recovery of Phage Antibody from LCM-procured Tissue Pieces-Genes encoding scFv fragments were amplified by PCR from LCMprocured tumor pieces using the following primer pairs: Fd2 (TTTTT-GGAGATTTTCAAC) and Fdseq (GAATTTTCTGTATGAGG). The amplified fragments were digested by SfiI and NotI, purified, and ligated into Fd-Tet vectors precut with the same restriction enzymes (1). The ligation products were used to transform chemically competent TG1. Each LCM library contained Ͼ10 5 independent clones. The number of unique phage antibodies was determined by patterns of BstNI digestion (1,14). When restriction digestion patterns showed ambiguity, phage antibody genes were sequenced to determine their uniqueness.
Initial Analysis of Selection Output by FACS-Prostate cancer (PC3 and Du-145) or non-tumorigenic control (BPH-1) cells were incubated with phage antibody (5 ϫ 10 11 cfu/ml) for 1 h at 4°C. Bound phages were detected by FACS (LSRII, BD Biosciences) using biotinylated anti-M13 antibody (Sigma, diluted 1:1000) followed by streptavidinphycoerythrin (BIOSOURCE/Invitrogen, diluted 1:1000) (1). Phage antibodies that showed positive binding were identified and sequenced. Further Analysis of Selection Output by Immunohistochemistry-Sections of prostate cancer tissue (frozen and paraffin-embedded) and normal human tissues were provided by the Genitourinary Tissue Core of the University of California, San Francisco Comprehensive Cancer Center. All tissues were collected with consent at the Core using protocols approved by the Committee on Human Research. For immunohistochemical analysis, tissue sections were incubated with biotinylated, monomeric scFv (50 g/ml) at room temperature for 1 h, washed with PBS, and incubated with horseradish peroxidase-conjugated streptavidin at a dilution of 1:1000 (Sigma) for 30 min. Binding was detected using diaminobenzidine (DAB) as the substrate (Sigma) (1).
Expression, Purification, and Biotinylation of scFv Fragments-Two forms of soluble antibody fragments, scFv and (scFvЈ) 2 , were produced (1). The scFv gene was subcloned into the secretion vector pUC119mycHis, adding a c-Myc epitope tag and hexahistidine tag at the C terminus of the scFv (1). To create the (scFvЈ) 2 dimer for immunoliposome studies, the c-Myc epitope tag was removed, and a free cysteine was introduced at the C terminus of the scFv preceding the hexahistidine tag as described previously (1). scFv monomer or (scFvЈ) 2 dimer proteins were harvested from the bacterial periplasmic space and purified by IMAC as described previously (1). To biotinylate scFv for FACS analysis, affinity-captured monomeric scFv fragments were washed in PBS and incubated with NHS-LC-biotin (Pierce) at 0.5 mg/ml for 20 min prior to elution with 250 mM imidazole.

Selection of Phage Antibody Binding to Clinically Relevant
Internalizing Epitopes by LCM-Selection was performed according to the scheme outlined in Fig. 1. We aimed to identify phage antibodies that bind to tumor epitopes present on actual cases of cancer and to further identify a subset of functional phage antibodies that bind to internalizing epitopes so that they may be exploited to deliver payload to the interior of tumor cells.
We devised a multistep strategy to achieve these aims (Fig.  1). First, a sublibrary was generated that is enriched for binders to cell surface receptors including those that are internalizing. This was accomplished by counterselecting a naïve phage antibody library containing 5 ϫ 10 8 unique scFv frag-ments on a panel of non-tumorigenic epithelial cell lines to remove binders to common cell surface antigens followed by selecting on a panel of live tumor cell lines such as the hormone refractory prostate cancer lines PC3 and Du-145 (1,14,34,35). By manipulating the selection conditions to preferentially recover internalized phage, a sublibrary enriched for internalizing phage antibody was created (1,20,25,36). Next the enriched sublibrary was incubated with tumor tissue slides, and tumor cells along with bound phage were procured by LCM (Fig. 2). The scFv genes were amplified by PCR and recloned into a phage display vector to generate a population of phage antibody that were either screened or used as input for the next round of selection (Fig. 2). Following one or two rounds of selection on tissue, the output was screened first on tumor cell lines to identify positive binders. Following sequencing, unique scFv fragments were further studied by IHC on tissue slides according to the scheme outlined in Fig.  1. This selection scheme effectively restricts selection outcomes to phage antibodies that bind to epitopes present on both tumor cell lines and tumor cells in situ from actual cases. Moreover these antibodies possess internalizing functions that can be exploited for targeted payload delivery. Antibodies that meet these criteria will likely have significant therapeutic values.
Initial Analysis of Selection Output: Binding to Tumor Cell Lines-Random clones from the sublibraries created after LCM-based selections were screened on PC3 and Du-145 cells by FACS (Fig. 3). More than 600 clones from various LCM-derived sublibraries were screened. Only those clones that bound to both PC3 and Du-145 cells were chosen for further analysis because they are more likely to recognize tumor cell surface antigens as opposed to artifacts associated with a particular tissue slide. The fraction of CaP cell linebinding clones ranged from 15 to 88% (Table I). Unique clones were identified by DNA sequencing. Thirteen unique phage antibodies were found from a total of 85 positive clones sequenced (Supplemental Table 1). We focused on two scFv fragments, UA20 and 585II41, for further characterization. The UA20 scFv was obtained from selection on paraffin-embedded prostate cancer tissue. The 585II41 scFv was obtained from selection on fresh frozen prostate cancer tissue.
Further Analysis of Selection Output: Binding to Tumor Cells in Situ-Phage antibodies selected by LCM were expected to bind to clinically relevant antigens on cancer cells in situ. We performed immunohistochemical studies using soluble scFv fragments derived from LCM-selected phage antibody on prostate cancer tissue sections. Fig. 4 shows the staining results of the UA20 and 585II41 scFv fragments on tissue specimens obtained from Gleason 3 ϩ 4 patients. On both frozen and paraffin-embedded tissue slides, the UA20 scFv showed an intense staining of tumor epithelium with minimal staining of normal adjacent prostate epithelium (Fig.  4, A and C). The 585II41 scFv also stained tumor cells intensely on frozen tissue slides (Fig. 4B). Some basal cells in normal epithelium adjacent to tumor were also stained with reduced intensity (Table II). The 585II41 scFv did not stain paraffin-embedded slides (data not shown), consistent with the fact that it was originally identified from selection on frozen tissue slides. These experiments indicate that antibodies obtained from LCM selection bind to antigens that exist in patient specimens and thus are clinically relevant to human prostate cancer. The corresponding antigens are likely targets for therapeutic intervention.
Tissue Specificity-To study the cross-reactivity of scFv fragments with normal tissues, we performed IHC studies on a panel of normal frozen human tissues using purified 585II41 and UA20 scFv fragments (Table II). Compared with controls, the 585II41 scFv showed no significant staining on most normal tissues studied, including the brain, kidney, and heart. There was, however, significant staining of bronchial epithelial cells and skin eccrines. The UA20 scFv, on the other hand, showed a more restricted staining pattern. At the concentration tested (50 g/ml), the UA20 scFv showed strong staining on prostate cancer tissues but no significant staining on the panel of normal tissues studied (Table II). We conclude that both scFv fragments recognize tumor cells in situ, and the UA20 scFv has very low cross-reactivity to normal human tissues.
Internalization and Payload Delivery to Prostate Cancer Cells-Phage antibodies selected by LCM were derived from a phage population that was panned on tumor cell lines using a functional selection process targeting receptor-mediated endocytosis. To confirm that selected phage antibodies possessed this phenotype and were endocytosed by CaP cells, the UA20 scFvЈ with a free cysteine at the C terminus was produced and conjugated to maleimide-activated liposomes containing a fluorescent probe, DiIC18(3)-DS, and incubated with BPH-1 (control), PC3, and Du-145 cells. These immunoliposomes were efficiently endocytosed by both PC3 and Du-145 cells (Fig. 5) with minimal uptake into BPH-1 cells (Fig.  5). Without conjugated scFv fragments, untargeted liposomes were not taken up by prostate cancer cells (Fig. 5C). Like the UA20 scFv-ILs, the 585II41-targeted liposomes were also efficiently taken up by prostate cancer cells (PC3 and Du-145) (data not shown). These experiments demonstrate that scFv antibodies selected by LCM retain internalizing functions and are capable of mediating efficient and specific payload delivery. These antibodies are candidates for the development of targeted therapeutics against prostate cancer.
Identification of ALCAM as a Tumor Antigen-The 585II41 scFv was sequenced and found to be highly homologous to a previously identified scFv, H3. These two scFv fragments differ by only two amino acids, none of which are in the CDR3 region that is critical for antigen binding (data not shown). The antigen recognized by the H3 scFv has been identified previously by us as ALCAM, also known as MEMD or CD166. 2 We hypothesized that the 585II41 scFv is a variant of the H3 scFv and binds to ALCAM. To test this hypothesis, we performed competition experiments using both H3 scFv and IgG to compete with the 585II41 scFv for binding to prostate cancer cells (Du-145). As controls, an scFv and its corresponding IgG that do not bind to ALCAM-expressing cells were included in the experiment. FACS analysis showed that both H3 scFv and IgG competed away binding by 585II41 scFv, whereas the control scFv and IgG did not (Fig. 6A). This suggests that the H3 scFv and the 585II41 scFv target the same antigen, i.e. ALCAM.
To further confirm that 585II41 scFv binds to ALCAM, we used 585II41 scFv to immunoprecipitate (IP) its target antigen from prostate cancer cell lysates. We probed the IP product with a commercial monoclonal antibody raised against a unique ALCAM peptide (Fig. 6B). This anti-ALCAM mAb recognized the IP product of 585II41 scFv but not that of the control OA12 scFv, thus confirming that ALCAM is the antigen targeted by the 585II41 scFv. In agreement with our own IHC studies, ALCAM has been shown by others to be overexpressed in 86% of prostate cancer cases (37). The fact that we identified a binder to a validated prostate cancer marker indicates that our LCM-based selection method is indeed capable of identifying clinically relevant tumor antigens. We

TABLE I Summary of selection results
Four paraffin-embedded and two frozen CaP tissues were used in the selection. The sublibraries constructed from PCR products contained 2-8 ϫ 10 5 unique clones. Binders to both PC3 and Du-145 cell lines were identified from each sublibrary by FACS screening. Between 10 and 20 positive clones from each group were sequenced to identify unique clones. Thirteen unique clones were identified from a total of 85 clones sequenced.

Cases
Tissue slides Tumor grades (Gleason scores) No. clones in sublibrary Cell line binders are currently working to identify the antigen recognized by the UA20 scFv.

DISCUSSION
The success of targeted cancer therapy depends in part on the availability of a panel of targeting agents such as mAbs that recognize tumor cell surface antigens present in clinical specimens. Much work has been done to generate mAbs against cell lines derived from primary tumor. It has become evident, however, that when removed from their original tissue environment cultured tumor cells variably up-and down-regulate expression of cell surface molecules relative to primary tumor cells. It is challenging yet desirable to identify the overlapping surface epitope space between tumor cell lines and tumor cells in actual cases. We developed an LCM-based strategy that allows the selection of phage antibody against tumor cells in situ within their proper stromal microenvironment. By preselecting a naïve phage antibody library on a panel of tumor cell lines under internalizing conditions, we created a sublibrary that is enriched for binders to functional cell surface epitopes. This sublibrary was then used for further selection on tissue slides. By precisely procuring tumor cells along with bound phage by LCM, we identified phage antibodies that bind to clinically represented tumor antigens. These antibodies meet the following criteria: 1) binding to internalizing cell surface epitopes present on tumor cell lines and 2) binding to epitopes present on tumor cells in situ. The ability to deliver payload intracellularly to target cells present in actual cases of human cancer makes these antibodies attractive candidates for therapeutic development.
We identified ALCAM, also known as MEMD or CD166, as the target for one of the selected antibodies. ALCAM, a member of the immunoglobulin superfamily, was originally shown to be overexpressed on highly metastatic melanoma cells (38). Recently it has been shown to be overexpressed in prostate carcinomas and to be predictive of prostate-specific antigen relapse (37). The fact that we found an scFv targeting a validated prostate cancer maker demonstrates the effectiveness of our approach.
ALCAM has also been identified by selecting a phage antibody library on an ovarian tumor cell line, and an immunotoxin has been made using the anti-ALCAM scFv (39). As this study dealt with cell line selection only, future IHC study will help determine whether ALCAM is indeed a marker for ovarian cancer. Therapies targeting ALCAM should also take into consideration its distribution on normal tissues as our IHC study showed that ALCAM is expressed on normal bronchial epithelial cells.
The sublibraries that were used for the LCM-based selection were generated from selection on tumor cell lines following counterselection on a panel of non-tumorigenic cell lines. As no cell lines are truly normal, it is possible that these non-tumorigenic cell lines share some surface antigens with tumor cells. To account for this possibility and to preserve antigens that are overexpressed, if not exclusively expressed, by tumor cells, we performed a moderate counterselection. We aimed to reduce binders to the most common cell surface antigens but not to eliminate all binders that cross-react with non-tumorigenic cell lines. The issues of tumor specificity and clinical relevance were addressed by direct selection and analysis on tissue sections instead.
We found some unexpected features associated with the LCM-based selection that may have hindered the application of LCM in phage antibody display. Most curiously, phage bound to LCM-procured tissue pieces seemingly lose their ability to infect bacteria, posing a challenge to library selection. We had initially sought to recover bound phage by standard methods, i.e. elution of phage with high pH buffer followed by neutralization and infection of TG1 bacterial cells (30,31). However, little bacterial FIG. 4. Immunohistochemistry studies. Biotinylated scFv fragments were used to stain CaP tissues. The UA20 scFv was originally isolated from selection on paraffin-embedded tissues; it stained tumor cells in both frozen and paraffin-embedded tissue slides. The 585II41 scFv was originally isolated from selection on frozen tissues; it stained tumor cells in frozen but not paraffin-embedded tissue slides. A, staining of frozen tissues with UA20 scFv. B, staining of frozen tissues with 585II41 scFv. C, staining of paraffin-embedded CaP tissues with UA20 scFv. growth was observed under various culture conditions (data not shown). This phenomenon was seen even in manually dissected tissue pieces that were not exposed to the UV laser used in the Leica LMD system (data not shown). Exposure to ethanol during slide preparation for LCM seems to be a factor contributing to the observed reduction in phage viability. Regardless of the cause, we circumvented this problem by using the genomes of phages bound to the procured cancer cell pieces as templates The numbers of cases studied are indicated. Biotinylated scFv fragments were first tested on cell lines to ensure binding activity and then used for IHC studies. As a control, a random scFv with no binding activity to cell lines was used to register the background level of staining. No change, no change in staining level was observed when compared with the result of the control scFv.
for amplification of scFv genes by PCR.
We identified 13 unique phage antibodies after sequencing 85 tumor-reactive clones. Because the sample size was small, it was not possible to predict the total number of unique clones in the selection output. Determining population diversity based on limited sample size is a complex statistical problem that cannot be solved by simple extrapolation (40,41).
Although LCM has the capacity to procure a single cell, we generally opted to procure a group of 20 -50 tumor cells for phage antibody selection. We found that it was rather difficult to recover phage antibodies from single cell procurement even by PCR amplification. In the rare cases that the phage antibodies were recovered, the diversity of scFv fragments was very low (in two of three cases, only a single unique clone was found among the 20 sequenced). Either the UV laser path encircling the single cell came too close to the bound phage, thereby damaging its DNA and reducing its viability for recovery, or there may be less than one recoverable phage bound per cell on tissue slides. In any event, we found that it was practical to procure 20 -50 cells at a time for phage selection. When a large cluster of topologically contiguous tumor cells cannot be found, we generally procured several small threeto five-cell clusters for analysis.
In the future, we envision the creation of a generic sublibrary that contains binders to a broad spectrum of cell surface antigens. This can be done by selecting the naïve phage display library on a large panel of existing tumor cell lines such as NCI 60 (42,43). This sublibrary can then be used as a universal input for LCM-based selection on tissues. Given the amount of paraffin-embedded and frozen tissues already archived, we anticipate the discovery of increasing numbers of functional epitopes present in actual cases of cancer.
Acknowledgments-We thank Dr. James D. Marks for naïve phage antibody library, Dr. Daryl C. Drummond and Dr. Audrey Roth for help FIG. 6. Identification of ALCAM/ MEMD/CD166 as the target of the 585-II41 scFv. A, binding of the 585II41 scFv to prostate cancer cells was specifically competed by a previously identified anti-ALCAM scFv, H3, and its corresponding IgG1 but not by a control scFv, OA12, and its corresponding IgG1. B, analysis of IP products by Western blot. Lysates from biotin surface-labeled Du-145 cells were incubated with 585II41 scFv and OA12 scFv (control) to generate IP products that were analyzed by Western blot using an ALCAM-specific commercial monoclonal antibody. Only the 585II41 scFv IP product reacted with the anti-ALCAM mAb. The band (indicated by an arrow) is located between 100 and 110 kDa. ALCAM is predicted to be a 65-kDa protein, but glycosylation causes it to appear as a band of ϳ105 kDa on SDS-PAGE, consistent with previous reports (44). MFI, mean fluorescence intensity.
with liposome studies, and Kevin Chew for help with immunohistochemistry experiments.
* This work was supported by NCI Grant R01 CA118919 and NIDDK Grant R21 DK066429 from the National Institutes of Health, United States Army Medical Research and Material Command Grant W81XWH0510027, and CaPCure (Prostate Cancer Research Foundation) (to B. L.). 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.