Glycoprotein NMB (GPNMB), a transmembrane glycoprotein highly expressed in high-grade gliomas (HGGs), is an attractive target in cancer immunotherapy. We isolated a GPNMB-specific scFv clone, G49, from a human synthetic phage-display library. To obtain mutant single-chain variable-fragment antibodies (scFvs) with improved affinity and immunotoxins with increased activity, we subjected G49 to in vitro affinity maturation by a complementarity-determining-region (CDR) random-mutagenesis technique. Using light-chain CDR3 mutagenesis, cell-based panning by phage display, subsequent heavy-chain CDR1 mutagenesis, and flow-cytometric selection by yeast-surface display, we generated the mutant scFv clone 902V, with an overall 11-fold increase in affinity for GPNMB. Clone 902V was further randomized throughout the whole scFv by error-prone PCR, and one mutant, F6V, was selected by yeast-surface display. F6V scFv, differing from 902V by one amino-acid change in the light-chain CDR2, exhibited an affinity for GPNMB of 0.30 nM. The F6V mutant scFv clone was fused with a truncated form of Pseudomonas exotoxin A to form the immunotoxin F6V-PE38. F6V-PE38 demonstrated significant protein-synthesis-inhibition activity on GPNMB-expressing glioma and malignant melanoma cells (IC50 = 0.5 ng/ml [8 pM]), a 60-fold improvement over G49 activity, but no cytotoxicity on GPNMB-negative cells. Furthermore, F6V-PE38 exhibited significant antitumor activity against subcutaneous malignant glioma xenografts in two nude-mouse models and a melanoma neoplastic meningitis model in athymic rats. These GPNMB-specific scFv antibodies and immunotoxins hold promise as reagents in targeted therapy for HGGs and other GPNMB-expressing malignancies.
Glioblastoma multiforme (GBM), the most common primary brain tumor in adults,1 is a lethal disease refractory to conventional therapies. In spite of advances in surgery and radiotherapy and the introduction of chemotherapeutic protocols, the mean postdiagnostic survival period for GBM patients remains at less than 1 year.2 It is thus imperative that new therapeutic approaches to this malignancy be developed.
Recently, immunotherapeutic approaches for cancer therapy, by which monoclonal antibodies (Mabs) target tumor-specific antigens, have shown great potential.3 Mabs that recognize a wide variety of cell-surface targets on glioma cells have been constructed and have been applied to the treatment of malignant gliomas.4 Glioma-associated antigens targeted by immunotherapeutic approaches include cell adhesion molecules, matrix proteins,5 and growth factor receptors6 and their glioma-associated variants.7 Although these tumor-associated antigens are overexpressed in tumors relative to normal brain tissues, one problem associated with immune-based therapy for GBM is the heterogeneity of antigen molecules expressed on tumor cells.8 So far, although overlapping gene expression profiles are seen among different GBM cases, no single gene product is overexpressed in all GBMs. This limitation can be overcome by expanding the panel of GBM-specific targetable molecules and by using a combination of immunotoxins against different antigens9 that is optimized for each patient.
To generate Mabs that have high affinity for GMB-specific antigens, a phage display and yeast surface display system has been found to be a promising alternative to hybridoma-based Mab production technology.10, 11 Improved selection techniques for surface display of variable regions of antibody fragments have made it possible to directly isolate specific, high-affinity single-chain antibody fragments (scFvs)—the VH and VL immunoglobulin domains connected by a flexible linker—from a diverse antibody gene pool. The selected scFv units can be engineered to a therapeutic reagent for directing natural toxins12 and radioactive compounds7 to antigen-positive cancer cells. Compared to the intact immunoglobulin molecules, the smaller scFvs have the potential for better tumor penetration13 and rapid clearance from the blood pool and normal tissues.
One type of such an engineered, recombinant immunotoxin for cancer therapy is a chimeric protein that consists of a targeting moiety linked to catalytic domains of a natural toxin.14Pseudomonas exotoxin A is a potent bacterial toxin composed of three domains15: Domain I binds to the cell-surface α2 macroglobulin receptor, which mediates the internalization of the toxin;16 domain II is necessary for translocation of the toxin to cytosol; domain III inactivates elongation factor 2 by ADP-ribosylation, leading to arrest of protein synthesis and programmed cell death. Generally, the variable region of specific antibody replaces the cell-binding domain I to direct the toxin to target antigens that are expressed on the surface of cancer cells. Recombinant immunotoxins have been used for patients with lymphoma, leukemia and breast cancer.17–19
Increasing the affinity of an scFv improves the antitumor activity of recombinant immunotoxins. Several methods based on phage display technology have been developed to select higher-affinity antibodies.20–24 However, these strategies depend on the use of huge libraries that are often difficult to construct and maintain. Recently, though, scFv antibodies with increased affinity and immunotoxins with improved activity have been obtained from relatively small phage libraries by introducing mutations at specific positions, called hot spots.25–28 Hot spots are DNA sequences in the variable regions that are frequently mutated during affinity maturation in vivo. Among several different types of variable-region hot spots, the consensus hot spot sequence is represented by tetranucleotide Pu-G-Py-A/T.
Glycoprotein NMB (GPNMB) is a type I transmembrane glycoprotein that has recently been identified as a glioma-specific marker gene by the serial analysis of gene expression (SAGE) method.29 The predicted amino acid sequence indicates that GPNMB protein consists of three domains: an extracellular domain (ECD, 464 amino acids) preceded by a signal peptide, a single transmembrane region, and a relatively short cytoplasmic domain composed of 53 amino acid residues. Although the biological function of GPNMB remains to be seen, transfection of a minimally transformed human fetal astrocyte cell line with GPNMB cDNA resulted in a change in the phenotype of the tumor xenografts, from minimally invasive to highly invasive and metastatic.30 More recently, in our study of high-grade glioma (HGG) biopsy samples, GPNMB RNA transcripts were detected in 35/50 GBMs (70%), while little or no GPNMB mRNA expression was noted in normal brain samples. By immunohistochemical study of a larger HGG group, 75/108 GBMs (70%) were positive for GPNMB protein expression.31 Furthermore, quantitative flow cytometric analysis of fresh GBM biopsy specimens revealed that cell-surface GPNMB molecular density ranged from 1.1 to 7.8 × 104 molecules.31 These levels are sufficient for molecular targeting therapy by Mabs.32 Its frequent expression in human HGGs and its cell-surface localization make GPNMB a good target for antibody-mediated delivery of cytotoxic agents.
We here report the isolation of an anti-GPNMB scFv clone from a human synthetic phage display library. This scFv antibody binds to GPNMB with high affinity, reacts with GPNMB-expressing cells and demonstrates cell-killing activity against GPNMB-positive glioma cells when converted to an immunotoxin form. Furthermore, using a random mutagenesis technique in which mutations are introduced into complementarity-determining region 3 (CDR3) of the light chain and subsequently into the CDR1 of the heavy chain, we were able to isolate from relatively small libraries a mutant scFv that had an 11-fold increase in affinity. This mutant was further mutated in the whole scFv, and we have generated one scFv clone with an overall 28-fold increase in affinity for GPNMB. The immunotoxin constructed with this affinity-matured scFv exhibited significant improvement, as compared to intact immunoglobulin molecules, in cytotoxic activity on GPNMB-expressing glioma cells.
A human synthetic phage display library, Griffin.1 library, of 1.2 × 109 diversity, was obtained from Dr. Fiona Sait (MRC Center for Protein Engineering, Cambridge, UK), and phage members were rescued according to the provider's instructions.33
For use as a target antigen, recombinant protein defining the ECD of GPNMB (GPNMBECD, encoded by nucleotides 67–1458, excluding the signal peptide)34 was produced in Sf9 insect cells as described previously31 and biotinylated with Sulfo-NHS-LC-Biotin (Pierce, Rockford, IL). Panning was done in solution as described by Amersdorfer and Marks.35 The detailed panning procedure is presented in the Supporting Information Methods section.
Phage and scFv ELISA
The wells of a 96-well polyvinyl chloride microtiter plate were coated with 1 μg of GPNMBECD protein or appropriate control antigens in 200 mM NaHCO3, pH 9.6. The plates were then washed three times with 0.05% Tween/PBS and blocked with 2% milk/PBS. Phages (1 × 109 PFUs/well) or scFv antibodies diluted in 2% milk/PBS were added to wells and incubated for 1 h at room temperature and then washed three times as above. For phage enzyme-linked immunosorbent assay (ELISA), bound phages were detected with horseradish peroxidase (HRP)-conjugated anti-M13 antibody (1:5000, Pharmacia, Sweden). For scFv, HRP-conjugated anti-myc antibody (1:5000, Invitrogen, Carlsbad, CA) targeting the myc epitope present at the carboxy terminus was used for detection of bound scFv antibodies.
Preparation of scFv antibody
DNA fragment material encoding scFv was excised from the corresponding phagemid by NcoI and NotI digestion and ligated into the NcoI-NotI sites of pET22 vector (Novagen, Madison, WI), in which scFv protein was tagged at the carboxy terminus with the hexahistidine and myc sequences for purification and detection. Resulting plasmids were introduced to Escherichia coli BL21(DE3) cells (Stratagene, La Jolla, CA), and scFv antibody was expressed and purified from the soluble cytoplasmic fraction of bacterial lysate by using metal affinity resin (BD Talon, BD Biosciences, Palo Alto, CA) according to the manufacturer's instructions.
Surface plasmon resonance
Binding kinetic profiles of purified anti-GPNMB scFv antibody and immunotoxins were measured by surface plasmon resonance by using a Biacore 3000 biosensor system (Biacore Life Sciences, GE Healthcare, Little Chalfont, Buckinghamshire, UK). As an antigen, GPNMBECD protein was immobilized on the surface of the CM5 sensor chip at pH 5.5. Test samples were diluted in running buffer (10 mM HEPES/150 mM NaCl/3.4 mM EDTA, pH 7.4) and passed over the chip at concentrations ranging from 25 to 200 nM. The association and dissociation rate constants and the average affinity were determined by using the nonlinear curve-fitting BIAevaluation software (Biacore).
Indirect fluorescence-activated cell sorting (FACS) analysis was performed as previously described.31, 36 To prevent internalization of target antigens during assays, all the reagents and buffers were kept on ice, and experiments were performed at 4°C. Briefly, 1 × 106 cells were suspended in 100 μl of PBS containing 5% normal goat serum (NGS; 5% NGS/PBS) and blocked for 20 min. After two washes with 5% NGS/PBS, cells were reacted with anti-GPNMB immunotoxins or negative control anti-Tac(Fv)-PE3837 at 10 μg/ml in 5% NGS-PBS for 1 h. After washing, cells were incubated with rabbit anti-Pseudomonas exotoxin A antibody (Sigma, St. Louis, MO), followed by labeling with FITC-conjugated goat anti-rabbit IgG antibody (Zymed, South San Francisco, CA). Stained cells were analyzed on a FACSort flow cytometer equipped with Lysys software (Becton, Dickinson and Company, San Jose, CA). The number of GPNMB molecules expressed per cell, by cell populations, was determined by quantitative FACS (QFACS) determination of receptor density using the Quantum Simply Cellular system (Bangs Laboratories, Fishers, IN) as described previously.36 The techniques for disaggregation of biopsy- and xenograft-derived cells and preparation for flow cytometry also have been thoroughly described.31
Preparation of recombinant immunotoxins
The parental G49 and mutant scFvs were used to generate immunotoxin by fusing with the sequences for domains II and III of Pseudomonas exotoxin A (PE38) or a variant carrying a C-terminal KDEL peptide for improved intracellular transport, according to a protocol described previously.17, 26 The toxin was further purified as a monomer (64 kDa) by ion exchange and size exclusion chromatography to greater than 95% purity as shown in Supporting Information Figure S1, and no dimer or aggregate was detected. Typically, the yields for the immunotoxin production were between 1% and 5%.
Activity of immunotoxins on cultured cell lines was assayed by inhibition of protein synthesis as described previously.27 Cells were plated in 96-well plates at 2 × 104 cells in 180 μl of complete medium per well 24 h before the assay. Immunotoxins were serially diluted (0.01–1000 ng/ml) in PBS containing 0.2% bovine serum albumin (BSA; 0.2% BSA/PBS), and 20 μl of diluted toxin was added to each well. Plates were incubated for 20 h at 37°C and then pulsed with 1 μCi/well of L-[4,5-3H]leucine (Amersham Biosciences, Little Chalfont, Buckinghamshire, UK) in 20 μl of 0.2% BSA/PBS for 2 h at 37°C. Radiolabeled cells were captured on filter-mats and counted in a MicroBeta scintillation counter (Perkin–Elmer, Shelton, CT). The activity of immunotoxin was defined by IC50, which was the toxin concentration that suppressed incorporation of radioactivity by 50% as compared to the cells that were not treated with toxin.
Cell viability assay
Discrimination of viable and nonviable cells after treatment with immunotoxin was done by using Guava ViaCount Reagent (Guava Technologies, Hayward, CA) according to the manufacturer's instructions.
Cell culture and transfection
Gliobl astoma-derived D392 MG, D54 MG, D245 MG and D212 MG cells and medulloblastoma-derived D487 MED cells were established in our laboratory.38 Glioma cell line U87, melanoma cell lines SK-MEL-28, SK-MEL-2 and WM39, and medulloblastoma cell line DAOY were obtained from the American Type Culture Collection (Manassas, VA). All glioma cells, SK-MEL-28, and DAOY were maintained in Zinc Option Medium supplemented with 10% fetal bovine serum (FBS). Melanoma cell line SK-MEL-2 was cultured in RPMI1640 medium with 10% FBS. WM39 cells were grown in a mixture of MCDB153 (Sigma) and Leibovitz's L-15 Medium (Invitrogen, Carlsbad, CA) (ratio 4:1) supplemented with 2 mM of glutamine, 0.005 mg/ml of bovine insulin, 1.68 mM of CaCl2, and 2% FBS. To generate stable transformant U87-GPNMB overexpressing GPNMB protein, retroviral transduction with full-length GPNMB was performed as described previously.31 Increased cell-surface-GPNMB protein expression in U87-GPNMB cells was confirmed by indirect FACS analysis using anti-GPNMB rabbit polyclonal antibody 264031 and anti-GPNMB Mab 6FG11 (data not shown).
Antitumor activity in nude mouse subcutaneous xenograft model
Pediatric GBM-derived D212 MG and medulloblastoma D487MED xenografts were established in athymic mice. The test mice were treated by intratumoral injection every other day, on days 2, 4 and 6 after tumor inoculation, with a total of three doses of F6V-PE38. The control mice were handled in the same manner and treated with 0.2% HSA/PBS or a counterpart immunotoxin anti-Tac-PE38 which is negative for GPNMB antigen. Tumors were measured twice weekly with a handheld vernier caliper, and the tumor volumes were calculated in cubic millimeters by using the formula: ([length] × [width2])/2. Animals were tested out of the study when tumor volume met both of the following criteria: (1) larger than 1000 mm3 and (2) five times its original treatment size. The response of the subcutaneous xenografts was assessed by delay in the growth of tumor in mice treated with drug as compared with growth in control mice (T–C). The growth delay was the difference between the median times required for tumors in treated (T) and control (C) mice to reach a volume that was five times the size at the initiation of therapy and >1000 mm3. Tumor regression was defined as a decrease in tumor volume over two successive measurements. Statistical analysis was performed using the Wilcoxon rank-sum test for growth delay and Fisher's exact test for tumor regressions, as previously described.39 The antitumor experiments were repeated twice.
Malignant melanoma neoplastic meningitis model in athymic rats
A malignant melanoma SK-MEL-28 intrathecal neoplastic meningitis model was created in athymic rats according to a method we have described previously.12 The untreated animals demonstrated a progressive loss of hindlimb motor function, followed by death. The median survival was ≈10 days. In the treatment group, animals (n = 10) were given three doses of 3 μg of F6V-PE38 or F6V-PE38KDEL immunotoxin diluted in 40 μl of 0.2% HSA/PBS on days 3, 5 and 7 (day of tumor cell injection = day 0). Control group animals (n = 10) were given 40 μl of anti-Tac-PE38 or anti-Tac-PE38KDEL on the same day.
Survival estimates were calculated by the Kaplan-Meier method. The Wilcoxon test was used to compare survival functions between groups. The Wilcoxon test places more weight on early survival times in assessing differences in survival between groups. A significant result (p ≤ 0.05) indicates that there is evidence to suggest that the groups being compared have significantly different survival curves. Median survival estimates from Wilcoxon tests comparing survival functions between groups were calculated.
Isolation of anti-GPNMB scFv clone G49 from human synthetic phage display library
To obtain GPNMB-specific scFv clones for use in targeted therapy, four rounds of panning with the human synthetic phage display library were performed on recombinant protein defining the GPNMBECD. Polyclonal phage ELISA using a pooled phage population rescued in each round of panning indicated enrichment of GPNMB binders after round 3 (data not shown). Twelve phage clones were randomly selected from round 3 and round 4 samples and assessed for GPNMB-binding ability by monoclonal phage ELISA. Positive reactivity (greater than threefold increase in absorbance over wild-type M13 phage) was observed in 5/12 phage clones (42%) from round 3 samples and 9/12 phage clones (75%) from round 4 samples. DNA sequencing revealed that all of these positive clones were identical in scFv nucleotide sequence, which indicated that panning of the naïve phage library with 109 members resulted in selection of one GPNMB-specific clone, which we designated as G49. The deduced amino acid sequence revealed that G49 scFv contained both the VH and VL chains from human germ-line immunoglobulin genes (Table 1).
Table 1. Stepwise in vitro-affinity maturation of anti-GPNMB scFvs by random mutagenesis
By ELISA, G49 scFv antibody purified to 95% homogeneity (Supporting Information Fig. S2) from bacterial lysate reacted with GPNMBECD protein in a dose-dependent manner, which indicates that selective GPNMB-binding activity of G49 phage is retained in the corresponding soluble scFv antibody (Supporting Information Fig. S2a,b). G49 scFv antibody had a KD for GPNMBECD protein of 8.4 nM, as measured by surface plasmon resonance (Biacore, Pharmacia, Sweden) (Supporting Information Table S2). We demonstrated the specificity of G49 scFv by FACS analysis on live cells, where G49 bound significantly to only GPNMB-expressing cell lines D54 MG and U251/GPNMB and not to control cells (HEK293), while control scFv antibody (anti-Tac) did not bind to either of these GPNMB-expressing cell lines (Supporting Information Fig. S2c). The G49 scFv was used to generate immunotoxin by fusing with the sequences for domains II and III of Pseudomonas exotoxin A (PE38) or a variant carrying a C-terminal KDEL peptide for improved intracellular transport, according to the protocol described previously.17, 26 Toxins were each further purified as a monomer (64 kDa) by ion exchange and size exclusion chromatography to greater than 95% purity. The G49-PE38 recombinant immunotoxin inhibited 50% of the protein synthesis at a concentration of 30 ng/ml on GPNMB-expressing D392 MG when cells were exposed to the immunotoxin for 24 h, while control anti-Tac(Fv)-PE38 immunotoxin showed no cytotoxic activity at concentrations up to 1000 ng/ml (Fig. 1a and Table 1). G49-PE38 was toxic to D54 MG glioma cells, with an IC50 of 120 ng/ml (Table 2). No cytotoxicity was noted on GPNMB-negative cell lines, including HEK293 (Fig. 1b), A431, and mouse fibroblast NR6, at concentrations up to 1000 ng/ml (data not shown), which indicates that cytotoxicity of G49-PE38 is restricted to GPNMB-expressing cells.
Table 2. Cytotoxicity of anti-GPNMB immunotoxins; improvement by affinity maturation
Affinity maturation of G49
To obtain mutant scFvs with improved affinity and immunotoxins with increased activity, we subjected G49 to in vitro affinity maturation by a CDR random mutagenesis technique. Using light-chain CDR3 mutagenesis, cell-based panning by phage display, subsequent heavy-chain CDR1 mutagenesis and flow-cytometric selection by yeast-surface display, we generated the mutant scFv clone 902V, with an overall 11-fold increase in affinity for GPNMB. Clone 902V was further randomized throughout the whole scFv by error-prone PCR, and one mutant, F6V, was selected by yeast-surface display. F6V scFv, differing from 902V by one amino-acid change in the light-chain CDR2, exhibited an affinity for GPNMB of 0.30 nM. Detailed results of in vitro affinity maturation are presented in Supporting Information Results.
Binding kinetics analysis
Binding affinities of the parental and mutant anti-GPNMB immunotoxins were measured by surface plasmon resonance (Biacore) using GPNMBECD protein produced in insect cells and immobilized on a Biacore chip. As shown in Supporting Information Table S2, we found that, compared to the original G49-PE38 toxin, L22-PE38 (VL CDR3 mutant) had a higher association rate without sacrificing the dissociation rate, resulting in about 2.5-fold improvement in overall affinity, with a KD of 3.7 nM. L22-PE38 was further affinity-matured for the VH CDR1 domain to obtain 902V-PE38 toxin. Compared to L22-PE38, 902V-PE38 exhibited a 2-fold and 2.3-fold improvement in the association rate and dissociation rate, respectively. The 902V-PE38 immunotoxin had a KD of 0.77 nM. The whole 902V scFv was then randomly mutagenized by error-prone PCR in order to screen for mutants with increased affinity. The F6V mutant immunotoxin, which has one amino acid change in the VH CDR2 domain compared to the 902V clone, showed improvement both in association rate and dissociation rate, resulting in a KD of 0.3 nM (Table 1 and Supporting Information Table S2).
Cell surface binding determined by flow cytometry
To determine whether the anti-GPNMB immunotoxin was able to bind to native GPNMB proteins expressed on the surface of tumor cells, indirect and quantitative flow cytometric analyses were performed using human glioma and melanoma cell lines that express cell surface GPNMB at different levels (Table 2).31 Indirect flow cytometric analysis revealed that the affinity-matured F6V-PE38 immunotoxin strongly reacted with the human glioma cell lines D392 MG and D54 MG, which express GPNMB at high levels (Fig. 2). Disappearance of a peak shift was observed by coincubation with 10-fold molar excess GPNMBECD protein, indicating that the F6V-PE38 toxin bound to D54 MG cells via the scFv domain. The F6V-PE38 reacted weakly with U87 cells, which express surface GPNMB at low levels, while a significant right shift of the fluorescent peak was noted in stable transformant U87-GPNMB cells, confirming the binding specificity of this toxin. The F6V-PE38 reacted with GPNMB-expressing human malignant melanoma cell lines WM39 and SK-MEL-28 as well.31, 40 The F6V-PE38 immunotoxin failed to bind to GPNMB-negative HEK293 cells (Fig. 2). These results demonstrate that the anti-GPNMB immunotoxin F6V-PE38 toxin is able to bind specifically to native GPNMB protein molecules expressed on the surface of human HGG tumor cells.
Activity of anti-GPNMB immunotoxins
The activity of anti-GPNMB immunotoxins (parental G49 and mutants after affinity maturation) was examined on GPNMB-expressing glioma cell lines. The parental G49-PE38 immunotoxin had IC50 values of 30 ng/ml and 120 ng/ml on D392 MG and D54 MG cells, respectively. In accordance with the increase in affinity, successive affinity maturation resulted in stepwise improvement in the GPNMB-dependent cytocidal activity. The L22 (VL CDR3), 902V (VL CDR3 + VH CDR1), and F6V (VLCDR3 + VHCDR1 + VLCDR2) mutant toxins exhibited improved activity on D392 MG cells, with IC50 values of 6 ng/ml, 1 ng/ml and 0.5 ng/ml, representing 5-, 30- and 60-fold improvement over the original G49 toxin, respectively (Fig. 1a and Table 2). With these affinity-matured toxins, similar levels of improvement in cytotoxic activity was noted on D54 MG glioma cells, whereas there was no significant effect on the inhibition of protein synthesis in GPNMB-negative cells, including HEK293, A431 and NR6, at concentrations up to 1000 ng/ml (Fig. 1b and Table 2). Coincubation with 50-fold molar excess of GPNMBECD protein abrogated the cytotoxicity of F6V-PE38 on D54 MG (Fig. 1c), which indicates that the cell-killing activity of anti-GPNMB toxin is dose-dependent on the specific interaction of the scFv domain with cell-surface receptor molecules. The cell viability assays also showed that F6V-PE38 caused D392 MG cell death after incubation for 24 h in a dose-dependent manner, but had no effect on HEK293 cells (Fig. 1d). The stable transfectant U87-GPNMB, expressing cell-surface GPNMB at high levels, was generated by retroviral transduction of the glioma cell line U87, which expresses surface GPNMB at minimal levels (Fig. 2). As shown in Table 2, overexpression of surface GPNMB conferred significant sensitivity to the F6V-PE38 toxin, with an IC50 of 8 ng/ml, which supports the GPNMB-specific cytotoxic activity of this toxin. Taken together, our results demonstrate that the affinity-matured anti-GPNMB immunotoxin F6V-PE38 exhibited significant cytotoxic activity (IC50 ranging from 0.5 to 50 ng/ml) on GPNMB-expressing malignant glioma (D293 MG, D54 MG, D245 MG and D212 MG), malignant melanoma (WM39, SK-MEL28 and SK-MEL-2), and medulloblastoma (D487MED and DAOY) cells (Table 2).
Antitumor activity of F6V-PE38 in nude mouse subcutaneous xenograft models
To evaluate the antitumor activity of F6V-PE38, animals bearing D212 MG and D487 MED were treated intratumorally with three doses of 5 μg of F6V immunotoxins. An antitumor activity study was conducted in D212 MG with two concentrations of F6V-PE38. Relative to PBS-treated controls, tumors in mice treated with 5 μg and 10 μg of F6V-PE38, demonstrated statistically significant growth delay in a dose-dependent manner (Table 3). F6V-PE38 at 5 μg yielded a benefit of 17.6 days, with 8 of 10 tumors regressing, and the 10-μg cohort yielded a benefit of 23.8 days, with 7 of 10 tumors having regressed. In a similar pattern of delayed growth (T–C), the counterpart immunotoxin (anti-Tac-PE38) in the cohort treated with F6V-PE38 at 5 μg yielded a benefit of 10.9 days, and the 10-μg cohort yielded a benefit of 17.1 days. In addition to this study, we also tested F6V-PE38 against D487MED and demonstrated a beneficial effect. Compared to PBS-treated controls, F6V-PE38 yielded a significant T–C of 23.4 days, along with regression in 7 of 9 tumors (Table 3).
Table 3. In vivo antitumor activity of F6V-PE38 on D212 MG and D487 MED xenografts
Treatment of malignant melanoma neoplastic meningitis in athymic rats
A melanoma neoplastic meningitis model was also created in athymic rats12 from athymic mouse xenografts of malignant melanoma SK-MEL-28. Untreated animals demonstrated a progressive loss of hindlimb motor function, followed by death. In treatment groups, animals (n = 10) were given three doses, intrathecally, of 3 μg of F6V-PE38 immunotoxin diluted in 40 μl of saline/HSA on days 3, 5 and 7 (day of tumor cell injection = day 0), for a total dose of 9 μg. A control group (n = 10) was given, intrathecally, 40 μl of the same dose of anti-Tac-PE38 immunotoxin by the same treatment schedule (Fig. 3a). The recombinant immunotoxin F6V-PE38 showed significant therapeutic effect against cells from human xenograft SK-MEL-28 growing in the spinal fluid of immunosuppressed rats. The median survival of treated animals was 7.5 days longer than that of controls (Fig. 3a). For the first experiment, there was a significant difference in overall survival between the control group [anti-Tac(Fv)-PE38] and the F6V-PE38 group (Wilcoxon p = 0.0498; Fig. 3 and Supporting Information Table S3). For a repeat experiment, there was a significant difference in overall survival between the control, F6V-PE38, and F6V-PE38KDEL groups (Wilcoxon p = 0.0107; Fig. 3 and Supporting Information Table S3). Controls were significantly different from F6V-PE38 (Wilcoxon p = 0.0188) and F6V-PE38KDEL (Wilcoxon p = 0.0239). The experimental groups, F6V-PE38 and F6V-PE38KDEL, however, were not significantly different (Wilcoxon p = 0.2068).
GPNMB, a type I transmembrane protein, is a promising target for immune-based therapy of malignant gliomas31 and melanomas.40 In an immunohistochemical study of HGG biopsy specimens, 70% of the GBM cases were positive for GPNMB protein expression, while no remarkable staining was noted in surrounding normal brain tissues.31 Moreover, by quantitative flow cytometric analysis, cell-surface GPNMB molecular density ranged from 1.1 to 7.8 × 104 molecules/cell in fresh GPNMB-positive GBM biopsy specimens,31 which indicated that a sufficient amount of surface GPNMB molecules is expressed in vivo for molecular targeting therapy by Mab.32
In addition to tumor-restricted distribution, membrane localization, and sufficient cell surface density, an ideal tumor-associated antigen in Mab-based therapy must fulfill several requirements.4 Efficient internalization upon binding of ligand or antibody is a prerequisite for tumor antigens in targeted therapy using immunotoxins. To exert cell-killing activity, a recombinant immunotoxin bound to a cell-surface antigen must be internalized and transported to the cytosol compartment, where the toxin moiety catalytically inhibits a critical cell function, which leads to cell death.17 We found that the GPNMB antigen and our tested anti-GPNMB agents fulfilled these requirements. In our study, affinity-matured anti-GPNMB immunotoxins demonstrated cytotoxic activity on GPNMB-expressing glioma cells with IC50 values comparable to those of the prototype recombinant immunotoxin targeting mutant epidermal growth factor receptor.41 In GPNMB-negative cells, the cytotoxicity of anti-GPNMB toxins was efficiently blocked by coincubation of antigen, and no toxicity was noted. The cytoplasmic domain of GPNMB contains a di-leucine-based motif and a tyrosine-based motif that are able to induce endocytosis of surface protein and delivery to lysosomes.42 Although the precise mechanisms of membrane trafficking and internalization of GPNMB protein molecules remain to be seen, the fact that anti-GPNMB immunotoxin causes receptor-dependent cell death indicates that endocytosis of GPNMB is likely a regular process. Cytotoxicity elicited by anti-GPNMB immunotoxin is much greater for glioma cells (D392 MG and D54 MG) than for melanoma cells (SK-MEL-2 and WM39), although both of these cell types possess the same levels of surface GPNMB molecules. This observation may be attributed to different internalization in glioma and melanoma cells.
One striking observation in our study is that in the VL CDR3 mutagenesis all of the mutants with increased affinity had amino acid substitutions at positions corresponding to a hot spot DNA sequence. Hot spots are DNA sequences in the variable regions that are frequently mutated during affinity maturation in vivo.43, 44 Among several different types of variable-region hot spots, the consensus hot spot sequence is represented by tetranucleotide Pu-G-Py-A/T. No VL CDR3 mutant clone that had an amino acid change outside the hot spot was obtained. Moreover, in randomization of the VH CDR3 domain, which has no consensus hot spot sequence, all of the selected scFvs were identical to the parental clone G49. These findings suggest that hot spot residues are able to tolerate mutations and also possess the potential to improve affinity. It has recently been shown that, by introducing mutations at specific hot spots, scFv antibodies with increased affinity and immunotoxins with improved activity can be obtained from relatively small phage libraries.25–28 The CDR random mutagenesis focusing on hot spot residues can minimize the size of a mutant library and facilitate the stepwise in vitro affinity maturation of scFv antibodies.
The selection of an scFv clone is a first step toward the construction of genetically engineered antibodies with improved in vivo performance, including optimal blood clearance and tumor penetration. The valence of an antibody fragment affects the functional affinity to cell-surface antigen. Either covalently or noncovalently linked, multivalent scFvs generally exhibit a higher functional affinity than monovalent forms, which results in better in vivo targeting. These GPNMB-specific scFvs can be used to direct radioisotopes or other cytotoxic drugs to GPNMB-expressing cancer cells and will be valuable reagents in the diagnosis and targeted therapy for HGG and other GPNMB-expressing malignancies. In the treatment of central nervous system malignancies, such immunotoxins are administered by convection-enhanced delivery to the brain tumor,45 which allows direct parenchymal infusion of therapeutics, bypassing the blood-brain barrier. Only trace amounts of immunotoxin may distribute systemically and the possibility of life-threatening side effects, such as lung edema, is minimal.
In conclusion, F6V-PE38 is an affinity-matured anti-GPNMB recombinant immunotoxin that targets malignant HGGs, medulloblastomas, and melanomas. The results presented here show that F6V-PE38 has significant activity due to its matured affinity for its target, GPNMB. F6V-PE38 shows significant activity in in vitro cell-killing assays and in vivo models of GPNMB-expressing xenografts. The specific anti-GPNMB activity of this new toxin may translate very well into the clinical setting. F6V-PE38 is likely to show increased efficacy in treating brain tumors and other GPNMB-expressing malignancies.
The authors thank Mr. Charles N. Pegram and Mr. Scott Szafranski for technical assistance in Biacore analysis and immunotoxin preparation. They also thank Ms. April Coan for in vivo antitumor survival statistical analyses. This research was supported in part by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.