|Home | About | Journals | Submit | Contact Us | Français|
Targeting angiogenesis is a promising approach to the treatment of solid tumors and age-related macular degeneration (AMD). Inhibition of vascularization has been validated by the successful marketing of monoclonal antibodies (mAbs) that target specific growth factors or their receptors, but there is considerable room for improvement in existing therapies. Combination of mAbs targeting both the VeGF and pDGF pathways has the potential to increase the efficacy of anti-angiogenic therapy without the accompanying toxicities of tyrosine kinase inhibitors and the inability to combine efficiently with traditional chemotherapeutics. However, development costs and regulatory issues have limited the use of combinatorial approaches for the generation of more efficacious treatments.
The concept of mediating disease pathology by targeting two antigens with one therapeutic was proposed over two decades ago. While mAbs are particularly suitable candidates for a dual-targeting approach, engineering bispecificity into one molecule can be difficult due to issues with expression and stability, which play a significant role in manufacturability. Here, we address these issues upstream in the process of developing a bispecific antibody (bsAb). Single-chain antibody fragments (scFvs) targeting pDGFRβ and VeGF-A were selected for superior stability. the scFvs were fused to both termini of human Fc to generate a bispecific, tetravalent molecule. resulting molecule displays potent activity, binds both targets simultaneously, and is stable in serum. assembly of a bsAb using stable monomeric units allowed development of an anti-pDGFRB/VeGF-A antibody capable of attenuating angiogenesis through two distinct pathways and represents an efficient method for rapid engineering of dual-targeting molecules.
Angiogenesis, the formation of sprouts from pre-existent blood vessels and their invasion into surrounding tissue, has been implicated in the development of cancer and age-related macular degeneration (AMD).1,2 Targeting the vascular endothelial growth factor (VEGF) pathway has provided significant benefit to patients with both cancer and AMD.3 However, efficacy is often moderate at best, and there is substantial room for improvement. Some patients show an intrinsic refractoriness to anti-VEGF therapy, while others show transient benefit followed by tumor progression.4–6 One mechanism of adaptive resistance to anti-VEGF-therapy in preclinical models involves pericytes that express platelet-derived growth factor receptor (PDGFR)β.7–9 Several lines of evidence support co-targeting the PDGF and VEGF pathways to enhance the efficacy of anti-angiogenic treatment. Co-inhibition of signaling mediated by PDGF/PDGFRβ and VEGF-A/VEGFR more effectively prevents the growth of new blood vessels and is superior to anti-VEGF therapy alone in inhibiting tumor growth in preclinical cancer models and in a mouse model of ocular neovascularization.10–14 A number of small molecule tyrosine kinase inhibitors (TKIs) have recently been approved for the treatment of cancer.15 These TKIs target a wide range of intracellular kinases, including VEGFR and PDGFR.16 However, treatment with these inhibitors is limited by toxicity, especially when combined with chemotherapy.17,18 An attractive approach to inhibition of angiogenesis in tumors is the use of a single protein therapeutic that interrupts both VEGF and PDGF pathways, and has the potential for superior anti-angiogenic activity over targeting VEGF alone. In addition, this approach is likely to result in lower toxicity than “broad spectrum” TKIs.
Bispecific antibodies (bsAbs) represent an intriguing approach to antibody engineering by combining two different binding specificities into one antibody-like molecule. The rationale for this approach was first described over two decades ago, and potential applications of bsAbs range from immunohistochemistry and diagnostics to human therapeutics.19,20 The promise of dual-targeting therapeutics has prompted the design of various formats of antibody-like proteins that can be tailored to specific needs.21,22 However, the manufacturability of these complex proteins at large scale continues to be a significant hurdle to the development of bsAbs.21,23–25 To date, no dual targeting proteins have been approved by the US Food and Drug Administration (FDA) for any indication.
Single-chain variable fragments (scFvs) are extremely useful in the construction of molecules with the complexity required for a dual-targeting approach. ScFvs contain both light (VL) and heavy (VH) variable domains in a single polypeptide, and can be fused to other proteins in modular form.26 The proximity of the VH and VL domains obviates the pairing of independent polypeptides, greatly facilitating production. In addition, the specificity, small size, short half-life in vivo, and potential for production from prokaryotic cells make scFvs attractive molecules for certain biomedical applications.20,27
Despite the advantages of scFvs, the development of these molecules has been limited by the prevalence of stability issues. Since the early reports on construction of scFvs, numerous groups have struggled with poor expression, insufficient affinity, aggregation, limited shelf-life and poor serum stability.25,28,29 Some of these issues have been addressed recently by applying techniques designed to enhance stability through protein engineering, resulting in a number of new drug candidates that demonstrate efficacy in clinical settings.30 However, the lack of FDA approval of a therapeutic molecule containing a scFv highlights the limitations of these proteins. As a result, attention continues to be focused on the folding and stability of scFvs.25
The observation that scFvs isolated from libraries of antibody fragments or assembled from IgGs frequently require structural modification to improve potency, stability, solubility or production has led to extensive examination of the factors that contribute to their inferior biophysical properties.31,32 In many cases, maturation techniques have been implemented to obtain the desired activity by mutation followed with rigorous functional screening.33–35 However, the theoretical diversity generated by combinatorial libraries of random or semi-random mutations can be very large. In order to sample a large population of variants, a display platform is typically used that maintains the link between a protein and the nucleic acid sequence of the encoding gene. This strategy of screening large numbers of candidates for the desired function, in combination with the evolution of molecular properties through mutation, is an alternative to rational or computational design, and circumvents the difficulty of predicting both the impact of mutations on structure and function of proteins and the greater diversity of options that can be explored.
Previously, we described a combinatorial method for selecting stable antibody fragments from a human phage display library and used the resulting scFvs to construct stable bispecific molecules against two soluble cytokines.36 In this report, we performed stress-guided selections for the isolation of stable antibody fragments targeting PDGFRβ and VEGF-A. The scFvs demonstrate high affinity binding to each target, high potency in cell-based assays, and remarkable stability in the absence of a traditional maturation campaign. The scFvs were fused to both ends of the Fc region of human IgG1 to construct a tetravalent bsAb. The final bispecific molecule binds both targets simultaneously, possesses the desired activity in cell-based assays, and is stable in vivo. Using a combination of analytical tools, we show that the stability of the final candidate is superior to that of extensively engineered antibody fragments. Furthermore, the dual-targeting molecule demonstrates the potential for production at large scale. The process we describe overcomes significant challenges in the development of this complex class of proteins, and generates a highly efficacious anti-angiogenic molecule with considerable therapeutic potential.
To identify antibody fragments with superior stability, scFvs were selected against soluble VEGF-A and PDGFRβ-Fc under conditions that incorporate incubation at elevated temperature (see Materials and Methods). We previously reported a strategy that converts antigen-binding fragments (Fabs) to scFvs in batch mode and incorporates a step that shuffles the variable regions to increase diversity.36 The conversion process was followed by additional rounds of panning that integrated thermal stress to select for stable scFvs that bind either VEGF-A or PDGFRβ and function as inhibitors of the respective pathways. The binding characteristics were determined for unique scFvs, and two fragments, one directed against VEGF-A and one directed against PDGFRβ, were selected for construction of a dual-targeting molecule. The biophysical properties of the component scFvs and the bispecific, IgG-like protein were characterized using several analytical tools: differential scanning calorimetry (DSC), size-exclusion chromatography combined with multi-angle light scattering (SEC-MALS), and dynamic light scattering (DLS). Collectively, these three techniques provide a thorough assessment of protein integrity.
DSC measures phase transitions in solution. Upon thermal treatment, changes in heat capacity are measured that correlate with protein unfolding or precipitation. The observed melting temperatures (Tms) represent the midpoint of protein unfolding where 50% of the molecules are folded and 50% unfolded. The two scFvs chosen for construction of the bispecific molecule demonstrated extraordinary stability with Tms of 74°C (anti-VEGFA scFv) and 78°C (anti-PDGFRβ scFv) (Fig. 1A). These are the highest reported values for scFvs of which we are aware, including those that have been subjected to maturation campaigns.37 The bispecific scFv-Fc-scFv constructed using these stable scFvs exhibits a primary Tm of 72°C, indicating that it too is highly stable (Fig. 1B). Similar Tms were obtained for the bispecific antibody and the component scFvs, which is consistent with our previous findings with dual-targeting molecules.36 The Tm derived for the bsAb is close to the Tm of the antibody fragment with the lowest melting point, the anti-PDGFRB scFv.
SEC-MALS was used to determine the mass and oligomerization state of the scFvs expressed in bacteria and the scFv-Fc-scFv expressed in CHO cells. Coupling static light scattering in-line to SEC allows for an absolute measurement of molecular weight (MW) that is independent of protein elution position. All three proteins were subjected to one freeze/thaw cycle prior to analysis by SEC-MALS and each was evaluated at a concentration of 1–2 mg/mL. The average MW of each scFv was consistent with the predicted values of 29 kD and 31 kD for the anti-VEGF-A and anti-PDGFRβ fragments, respectively. Both scFvs were present predominantly as monomeric species, with a small amount of higher MW species, including a dimeric form, which comprised approximately 10% for the anti-PDGFRβ and approximately 7% for the anti-VEGFA scFv (Fig. 2A and B). However, after fusion of the scFvs to the Fc domain to create the dual-targeting molecule, the percentage of higher MW species was diminished, with only 3% observed by SEC-MALS (Fig. 2C).
The third analytical tool applied to evaluate stability was DLS, a technique that is a very sensitive indicator of aggregation. DLS measures the fluctuations of light scattering intensity over a very short time scale that result from diffusion of molecules in solution. From these measurements, the hydrodynamic radius (Rh) is calculated. The two scFvs and the scFv-Fc-scFv, were analyzed at both low (1 mg/mL) and high (25 mg/mL) concentrations. As shown in Figure 3A and B, no large disordered aggregates were detected, and very little multimerization of the scFvs was observed at the lower concentration, consistent with the results obtained using SEC-MALS. However, the amount of larger multimers increased significantly at higher concentrations of the fragments, as evident in the more than doubling of the Rh. In contrast, the scFv-Fc-scFv exhibits a very similar profile at both high and low concentration (Fig. 3C) with a smaller increase in the average Rh, indicative of a slight increase in the presence of a small multimer. These results suggest that the aggregation potential of the scFvs is reduced when they are fused to the Fc domain.
The extraordinary stabilities observed for the two scFvs are likely the result of the rescue of particular amino acid sequences following selection under thermal stress. The variable region sequences were aligned with that of the germline families that most resemble the amino acid sequence of each scFv (Fig. 4).
Both antibody fragments belong to the same VH family due to the design of the Dyax library.38 VH3_3–23 was selected as the heavy chain framework because it possesses key amino acid residues in the framework region that have been reported to contribute to stability.32,38
One of the key issues with bispecific IgG-like molecules has been production. The DNA encoding the scFv-Fc-scFv was cloned into an expression vector for evaluation of production in CHO cells (as described in Materials and Methods). The bispecific molecule was produced at approximately 120 mg/L from an unoptimized pool under standard conditions, a level of production we typically observe with a native IgG antibody under similar conditions.
The affinities of all three proteins were measured using surface plasmon resonance (SPR). The anti-PDGFRβ scFv displayed impressive affinity for the PDGFRβ receptor with a calculated KD of 90 pM. The anti-VEGFA scFv also displayed sub-nanomolar affinity for the target ligand (KD of 600 pM). When fused to the Fc region of IgG to assemble the scFv-Fc-scFv bsAb, both arms exhibit sub-nanomolar affinities to each target (Table 1). However, there was an observed decrease in the association (on rate Ka) of binding of the bsAb to PDGFRβ, with no observed alteration in dissociation (off rate). The scFv-Fc-scFv construct was also evaluated for simultaneous binding to both targets. Figure 5 depicts the binding of the bsAb to immobilized VEGFA and the subsequent binding of PDGFRβ-Fc added in the final mobile phase. The association and dissociation curves generated for these interactions are consistent with binding curves generated against each target separately, suggesting that the bispecific molecule is capable of binding to both targets simultaneously without significant changes in affinity.
The anti-PDGFRβ scFv, the anti-VEGF-A scFv and the anti-PDGFRβ/VEGF-A scFv-Fc-scFv were evaluated for potency in cell-based assays that measure neutralization of the activity mediated through human VEGF-A and human PDGFRβ. In a short-term assay (15 minutes), the anti-VEGF-A scFv and the bispecific scFv-Fc-scFv neutralized VEGF-A-induced phosphorylation of VEGFR2 in HEK293 cells transfected with the receptor (IC50: 0.04 and 0.07 nM respectively, Suppl. Table 1). In a long-term assay (48 hours), the anti-VEGF-A scFv and the bsAb inhibited VEGF-A induced proliferation of human umbilical vein endothelial cells (HUVECs) with IC50 values in the picomolar range (Fig. 6A). The level of inhibition was comparable to that seen with bevacizumab (Fig. 6A). To test for neutralization of the PDGFRβ pathway, the ability to block the activity of PDGF-BB on human brain vascular pericytes (HBVPs) was analyzed. In a short-term assay (15 minutes), the anti-PDGFRβ scFv and the scFv-Fc-scFv blocked PDGF-BB-induced phosphorylation of PDGFRβ on HBVPs (IC50: 0.3 and 0.1 nM respectively, Suppl. Table 1). In a long-term assay (18 hours), the anti-PDGFRβ scFv and the scFv-Fc-scFv neutralized PDGF-BB-induced proliferation of HBVPs with picomolar IC50 values (Fig. 6B). These data suggest that the scFvs and the bispecific molecule are very potent inhibitors of activi tymediated by both targets. The decrease in observed on-rate of the scFv-Fc-scFv binding to PDGFRβ, relative to the anti-PDGFRβ scFv, had no impact on potency in the cell-based assay. In addition, anti-PDGFRβ/VEGF-A scFv-Fc-scFv did not inhibit the activity of mouse VEGF-A or activity mediated through mouse PDGFRβ (Suppl. Table 1).
Blockade of the VEGF and PDGF pathways together resulted in enhanced anti-angiogenic activity in mouse tumor models.13,14 To mimic this in vitro, a co-culture system using HUVECs and human mesenchymal stem cells (hMSCs) was developed to test inhibition of proliferation and sprouting of endothelial cells. It was reported previously that, when cultured together, endothelial cells induce differentiation of murine mesenchymal stem cells into pericytes.39 In the co-culture sprouting assay, hMSC and cytodex beads coated with HUVECs were embedded in fibrin gel to form endothelial sprouts, which are covered with pericytes that differentiate from hMSCs. Endothelial sprouting is dependent on the presence of VEGF-A in the media, whereas pericyte differentiation and coverage of endothelial cells is dependent on PDGF-BB (produced by endothelial cells) and PDGFRβ on the surface of the hMSCs and pericytes. Blocking the activity of VEGF-A at the start of the assay led to complete inhibition of endothelial sprouting, whereas blockade of activity mediated by PDGFRβ resulted in inhibition of pericyte differentiation and coverage of endothelial sprouts (data not shown). To mimic therapeutic inhibition of angiogenesis, antibodies or anti-PDGFRβ/VEGF-A scFv-Fc-scFv were added after sprouts and pericyte coverage had formed. Addition of the anti-VEGF-A mAb bevacizumab inhibited the formation of new sprouts, but had little effect on preformed sprouts (Fig. 7). Addition of anti-PDGFRβ mAb E9899 induced pericyte dissociation from endothelial cells, but had no effect on endothelial sprouting (data not shown). In contrast, addition of the scFv-Fc-scFv alone or of both anti-VEGF-A and anti-PDGFRβ antibodies together resulted in significant reduction of endothelial sprouting in this model and induced pericyte dissociation from endothelial cells (Fig. 7A and B). The effect on endothelial sprouting was dose dependent (Fig. 7C).
The pharmacokinetic properties of the anti-PDGFRβ/VEGF-A scFv-Fc-scFv were examined with a single intravenous injection in C.B-17 SCID mice. SCID mice were used for this experiment to understand better the behavior of the molecule in the strain used for tumor models. As shown in Table 2, a single injection of the scFv-Fc-Fv resulted in antibody-like clearance with a t1/2 of about 460 hours. Furthermore, sera isolated from these mice neutralized the activity of VEGF-A and PDGFRβ in the short-term activity assays (Suppl. Table 2). These data suggest that the dual-targeting molecule is not only stable in vivo, but also retains its functional properties in mouse serum for up to a week.
The potency of the anti-PDGFRβ/VEGF-A scFv-Fc-scFv was tested in vivo in a tumor model. Both scFvs in this engineered molecule bind specifically to the human protein and do not cross-react with mouse VEGF-A or mouse PDGFRβ. For this reason, the potency of both arms of the molecule could not be tested in a mouse model. However, a large number of human tumors secrete VEGF-A and growth of these tumors in mice has been shown to be dependent on production of the growth factor by the tumor. The A673 tumor model has been used extensively to study the potency of anti-human VEGF-A antibodies, including bevacizumab. This model was used to test the potency of the bsAb relative to bevacizumab. Prophylactic and therapeutic treatment with the scFv-Fc-scFv inhibited significantly the growth of A673 tumors in mice with efficacy similar to that seen with bevacizumab (Fig. 8A and B). Furthermore, treatment with the dual-targeting molecule or bevacizumab significantly decreased micro-vessel density within the tumors (Suppl. Fig. 1).
Classical antibodies to a number of cell surface receptors induce internalization of the receptor (and the antibody). The anti-PDGFRβ/anti-VEGF-A scFv-Fc-scFv was tested for internalization over time by incubating with HBVPs expressing endogenous human PDGFRβ. As shown in Figure 9, the scFv-Fc-scFv bound to the cell membrane at T0 (on ice) as evidenced by the plasma membrane staining. As early as thirty minutes after warming at 37°C, internalization was apparent as cytoplasmic punctuate staining. Internalization was monitored over time, and punctuate staining increased and then decreased, indicating the PDGFRβ/VEGF-A scFv-Fc-scFv was efficiently bound, and internalized by the HBVP. After two hours, almost no plasma membrane staining was visible, although in some cells hazy staining could be observed, possibly a result of constitutive receptor expression. Similar results were obtained with a mouse-anti-human PDGFRβ antibody E9899 (data not shown). Furthermore, VEGF-A bound to the VEGF-A-specific scFv of the dual-targeting entity did not inhibit internalization (data not shown).
Antibodies possess a modular structure that suggests numerous possibilities for the development of bispecific molecules mediating biological activity specific to each target. However, with the use of antibody fragments to construct IgG-like proteins capable of dual-targeting, stability issues tend to increase as the complexity of the molecules increases. The data in this report suggest that one effective strategy for the successful generation of bsAbs is to select stable components early in the development process. One of the V region pairs identified by stress-guided selection would not have been predicted to be exceptionally stable. The combination of VH3 and VκI has been reported to be one of the more stable V region pairs. However, the VH3/VκIV pairing was not noted to be particularly stable.32 One of the advantages of stress-guided selection is the identification of atypical V region pairings with the desired properties through functional analysis.
ScFvs can serve as versatile building blocks for the construction of bsAbs, but they typically require engineering or thorough characterization during the selection process due to stability issues with this class of proteins.25 Here, we incorporated stress-guided selection during panning of the library and chain shuffling during the conversion of the Fab output to scFvs in order to avoid the need for extensive maturation of individual antibody fragments. Our previous study indicated that shuffling of variable regions using a combinatorial strategy may enhance isolation of scFvs with high affinity for the target, and permits the construction of bsAbs with stable monomeric units.36 Thermal stress has been applied previously at both selection and screening steps of the process.25,40 Phage display is one of the few display systems in which thermal stress can be applied during selection. Thermal treatment prior to phage panning prevents the enrichment of unstable antibody fragments, while rescuing more stable fragments that either have the capacity to withstand unfolding or may refold upon cooling prior to selection against the target. In addition, thermal stress has been suggested to be superior to other stress-guided methods for maturation of individual fragments.40 However, stress during the selection process is directed toward proteins that are fused to the p3 coat protein of the phage, which may influence the stability of the displayed protein. The incorporation of thermal treatments in both the selection and screening processes would interrogate stability of fragments both as fusions and as soluble, independent proteins.
Another key element of the process is the incorporation of a combination of biophysical analyses to verify stability. While one analytical tool can provide insight into the biophysical behavior of a protein, the combination of multiple techniques provides comprehensive characterization during the selection of antibody fragments for assembly of complex, bispecific molecules. The anti-PDGFRβ/VEGF-A bsAb described here displays both potency and stability equivalent to typical therapeutic mAbs. The high MW species observed in preparations of the candidate scFvs are minimized when the single chain fragments are fused to an Fc domain. This effect is especially evident at higher concentrations, as determined by DLS and SEC-MALS. Results of analysis by DSC indicate that the thermal stability of less stable scFv is not altered by fusion to the Fc domain. The reduction in aggregation that is apparent upon evaluation using SEC-MALS and DLS may result from physical separation of aggregation-prone scFvs by anchoring them to the Fc domain. It is possible that the use of thermal stress during the selection of scFvs fused to p3 identifies antibody fragments that are stable when fused to Fc. Since the stability of the final molecule appears to be influenced significantly by the stabilities of the individual antibody fragments, as measured by DSC, our data suggest that the stability of a dual-targeting molecule in this format depends on the stabilities of the individual antibody units.
Stability of protein therapeutics has gained considerable attention due to the complexity of developing molecules with binding sites for different targets. However, confronting issues involving degradation, denaturation and aggregation downstream in the development process may not be the best approach. Engineering processes applied to “evolve” proteins by rational design late in the molecular development process are time consuming and labor intensive. On the other hand, the use of directed evolution to enhance stability is adaptable to high-throughput platforms, and can significantly reduce the time between discovery and lead development.
Targeting angiogenesis has been shown to be effective in the treatment of solid tumors and wet AMD.1 However, while inhibitors specific for VEGF-A have demonstrated efficacy, there is considerable room for improvement. In preclinical models, co-targeting the VEGF and PDGF pathways is more efficacious than targeting VEGF-A alone.11,13,14,41 VEGF-resistance in these preclinical models has been shown to be mediated by emergence of pericytes expressing PDGFRβ within tumors. Also, upregulation of PDGF-C has been demonstrated to play a role in resistance to anti-VEGF treatment both in preclinical and clinical settings.8,9,42,43 Thus, there is justification for the development of treatments that target both the VEGF and PDGF pathways. One successful approach is the use of multi-targeted TKIs that inhibit the intracellular kinase domains of both of these receptors.15 Although these molecules have been approved and successful in the clinic, their use is limited by toxicity, either as monotherapy or in combination with chemotherapy.18 As an alternative approach to the treatment of such tumors, a bispecific molecule that targets both the VEGF and PDGF pathways specifically offers the potential for effective inhibition of angiogenesis with reduced toxicity, relative to small molecule inhibitors, and may facilitate combination therapies with other drugs, including standard chemotherapy. This approach could provide more effective treatment for a variety of solid tumors.
The bsAb described in this manuscript is a high-affinity, dual-targeting protein specific for both PDGFRβ and VEGF-A. This scFv-Fc-scFv is effective in inhibiting VEGF-A and PDGFRβ activity in vitro (PDGF-BB, Fig. 6; PDGF-AB, CC, DD, Suppl. Table 1) and shows anti-tumor activity in vivo. The observed decrease in affinity of the bsAb to PDGFRβ, measured by SPR, did not appear to influence biological activity. We previously observed the changes in affinity from scFv to bispecific assembly that did not correlate with biological activity.36 Whether the changes in affinity are true or a result of steric hindrance related to Biacore formatting remains to be determined. Furthermore, the bsAb shows enhanced efficacy in an endothelial:pericyte co-culture assay, dependent on VEGF-A and PDGFRβ activity. These data support the use of this bsAb to block both pathways, and suggest that the molecule will provide improved efficacy in both oncology and wet AMD. Recent data with ranibizumab and a PDGF aptamer have shown encouraging anti-angiogenic effects in patients with wet AMD.44 However, this procedure involves two concurrent intravitreal injections at each dose. The engineered bsAb described here is expected to show enhanced efficacy in wet AMD with a single molecule. Furthermore, we anticipate reduced toxicity, either as a monotherapy or in combination with chemotherapeutics, relative to the less specific TKIs.
The method we have applied to engineering a dual-targeting molecule focuses on early identification of stable antibody fragments. As the next wave of protein-based therapeutics advance to lead development, the result of the successful application of engineering platforms, the value of applying engineering strategies and rigorous characterization upstream in the drug development process will become apparent.
Female C.B-17 SCID mice (CB17/Icr-Prkdcscid/IcrIcoCrl, Charles River Laboratories, Wilmington MA) were used for pharmacokinetic and tumor experiments. Bevacizumab (Genentech, San Francisco, CA) was obtained from the local Pharmacy. The rat anti-human PDGFRβ antibody (E9899) was generated by immunizing rats with soluble human PDGFRβ and selecting for neutralizing antibodies from hybridomas. Human umbilical vein endothelial cells (HUVEC) and primary human mesenchymal stem cells (HMSC) were obtained from Lonza (Walkersville, MD) and primary human brain vascular pericytes (HBVP) were purchased from ScienCell Research (San Diego, CA). Recombinant human VEGF-A165 was cloned, expressed and purified from E. coli at ZymoGenetics. Recombinant human PDGF-BB was generated in S. cerevisiae at Novo Nordisk (Copenhagen, Denmark) and provided to ZymoGenetics. A673 (CRL-1598) rhabdomyosarcoma was obtained from American Type Culture Collection (Manassas, VA). Human PDGFRβ-Fc, human VEGFR2-Fc, human VEGFA, human PDGF-BB and mouse anti-human PDGFRβ antibody were produced at ZymoGenetics. Monomeric PDGFRβ was prepared by a Lys-C digest of PDGFRβ-Fc, followed by affinity purification (anti-PDGFRβ sepharose). Biotin labeling of ligands was performed at ZymoGenetics.
Antibodies generated against both VEGF-A and PDGFRβ were derived from the Dyax libraries.38 The selections were performed as previously described with modifications.36 Anti-PDGFRB antibodies were identified by selecting on biotinylated target (in-house) captured on magnetic beads (Dynabeads M-280 Streptavidin, #112-06D, Invitrogen Dynal AS, Oslo, Norway). Anti-VEGF-A antibodies were identified by selecting on immunotubes (NUNC, Denmark) coated with antigen (VEGF-A in-house) at various concentrations. Following three rounds of selections, the Fabs in the enriched pool were converted to scFvs with shuffling of V regions through a combinatorial method.36 Additional rounds of panning were performed with the integration of thermal treatment (50–80°C, 1 hr) prior to incubation with target molecule. After 1–2 rounds of panning, scFvs were screened for activity using soluble scFv produced in E. coli as described previously.36
Anti-PDGFRβ clones were screened for antagonism using a blocking ELISA. Costar (#9018) 96-well plates were coated with an anti-human IgG antibody specific for Fcγ (#109-005-098, Jackson Immunology) in 0.1 M NaHCO3, pH 9.6 overnight at 4°C. The next day, plates were washed three times with 0.1% Tween-20/PBS (PBST) and blocked with 5% milk (#170-6404, Bio-Rad)/PBST for one hour at room temperature (RT). Next PDGFRβ was added at 0.25 µg/mL in 2% BSA (#160069 MB Biomedicals)/PBST and incubated for one hour at RT. Plates were washed and blocked again with 5% milk/PBST for one hour at RT. After another wash with PBST, a (1:1) mixture of supernatant containing either Fab or scFv and biotinylated PDGF-BB at 0.0112 µg/mL in 2% BSA/PBST was added for one hour at room temperature. Plates were washed with PBST followed by the addition of a 1:3,000 dilution of Streptavidin-HRP (#21124, Pierce) in 2% BSA/PBST for one hour at room temperature. Plates were then washed with PBST and 50 µL of TMB (TMBW-100 0-01, BioFX Laboratories) added. The color was allowed to develop for 20–30 min, followed by the addition of 50 µL of stop buffer (STPR-1000-01, BioFX Laboratories) to quench the reaction. Plates were then read at 450 nm on a plate reader.
Antibodies selected against VEGF-A were also screened for blocking the interaction between receptor and ligand. Costar (#9018) 96-well plates were coated with anti-human IgG Fcg-specific antibody (#109-005-098, Jackson Immunology) at 1 µg/mL in 0.1 M NaHCO3, pH 9.6 overnight at 4°C. The next day, plates were washed with 0.1% Tween-20/PBS (PBST) and blocked with 1% BSA (#A3059-100G, SIGMA)/PBST for one hour at room temperature (RT). Following a wash with PBST, VEGFR2-Fc at 0.2 µg/mL in 1% BSA/PBST was added and incubated for one hour at room temperature. Concurrently, in a separate 96 well plate (Costar 3357) a 1:1 dilution of supernatant, containing either Fab or scFv and biotinylated VEGF-A (ZymoGenetics) in 1% BSA/PBST at 20 ng/mL was made and incubated for 1 hr at RT. Blocked assay plates were washed with PBST followed by the addition of the supernatant/biotinylated VEGF-A complex for 1 hr at RT. After washing, a 1:4,000 dilution of Streptavidin-HRP (#21124, Pierce) in 1% BSA/PBST was added for one hour at RT. Plates were then washed and TMB (TMBW-1000-01, BioFX Laboratories) added to develop for 20 min, followed by the addition of 100 µL of stop buffer (STPR-1000-01, BioFX Laboratories) to quench the reaction. Plates were then read at 450 nm on a plate reader.
Selected scFvs were expressed using a proprietary vector with a phoA promoter in either BL21 (#69449-4, Novagen, Madison, WI) or TG1 cells (#200123, Stratagene, La Jolla, CA). Once transformed, colonies were selected for inoculation into 2 ml LB medium supplemented with antifoam at 100 µg/L (#A8311, Sigma-Aldrich, St. Louis, MO) and kanamycin (#60615, Sigma-Aldrich, St. Louis, MO) at 25 µg/mL. Cultures were grown overnight at 37°C, shaking at 250 rpm. The overnight cultures were then diluted to a 0.2% inoculum into 0.5 L of phosphate-limiting media45 and supplemented with antifoam and antibiotic as above. Cultures were grown in 2 L baffled flasks for 24 hrs at 30°C. Soluble scFv samples were recovered from wet cell pellets by treatment with periplasting buffer containing Ready-Lyse lysozyme (#R1802, Epicentre Biotechnologies, Madison, WI) as per manufacturer’s instructions in the Protocol for the Preparation of Periplasmic and Spheroplastic Proteins from >1 mL Bacterial Culture (Epicentre Biotechnologies).
For the Immobilized Metal Affinity Chromatography (IMAC) enrichment of scFvs, the periplasmic fraction was passed through a 0.22 µm filter and purified by affinity capture with a HisTrap HP column (GE Healthcare, Piscataway, NJ) on a liquid chromatography instrument (Akta Explorer System, GE Healthcare, Piscataway, NJ). The bound protein was eluted using 400 mM imidazole in 50 mM NaPO4, 500 mM NaCl pH 7.4 and assessed for protein content by absorbance at 280 nm, and for quality by analytical size exclusion chromatography and SDS-PAGE. For enrichment of Fabs, affinity purification was performed using Protein A (#17-1279-04, GE Healthcare, Piscataway, NJ) in batch mode. The Protein A resin was added to the periplasmic extract for overnight incubation at 4°C with mixing. The following day, the protein was eluted by a low pH step elution using 50 mM NaH2PO4 pH 2.5, 150 mM NaCl, immediately neutralized with 1 M HEPES pH 7.2 and evaluated as described above. Protein was dialyzed to a final buffer composition of 25 mM histidine, 125 mM NaCl, pH 6.8.
The scFv-Fc-scFv construct was assembled by a combination of PCR and homologous recombination in yeast, as an alternative to ligation, using a plasmid which was constructed from pZMP21 (patent pub. No. US 2003/0232414 A1) (deposited at the American Type Culture Collection, 10801 University Boulevard, Manassas, VA 20110-2209, designated as ATCC# PTA-5266) with DHFR as the selectable marker. The expression construct was electroporated into CHO DXB11 cells, adapted to grow in suspension in protein-free medium and subjected to nutrient selection in -HT medium followed by selection in methotrexate. Pools of selected cells were scaled up in spinner flask culture for purification and analysis of bispecific molecules. Conditioned media were harvested, passed through a 0.2 µm filter and adjusted to pH 7.4. The protein was purified from the filtered media using a combination of POROS® A50 Protein A Affinity Chromatography (Applied Biosciences, Foster City, CA) and Superdex 200 Size Exclusion Chromatography (GE Healthcare, Piscataway, NJ). Column fractions were analyzed by SDS-PAGE to determine the appropriate pools. Enriched protein was quantified by UV at A280 nm and an analytical SEC method was used to characterize the purified protein. Final buffer composition for the bsAb was equivalent to the individual scFvs (25 mM histidine, 125 mM NaCl, pH 6.8).
Differential scanning calorimetry (DSC) measurements were performed using a single cell VP-DSC instrument (MicroCal, Northampton, MA, USA) at a heating rate of 1°C/min for bispecific molecules and 1.5°C/min for single chain Fv. Working concentrations of these molecules were 0.35 mg/mL. Samples were buffer-exchanged over a Superdex 200 column, and a single peak of the target protein was collected, concentrated and filtered. The diluent was collected prior to injection of a bispecific molecule and was also used as reference blank for DSC analysis. All samples were stored frozen at −80°C until time of assay. Samples were thawed at room temperature, diluted to 0.35 mg/mL according to calculated extinction coefficient, and degassed prior to analysis.
Size Exclusion Chromatography Multi-Angle Light Scattering (SEC-MALS) analysis was used to provide an absolute measurement of MW and an assessment of oligomer formation from static light scattering in-line with SEC. Samples were analyzed with the three detectors connected in series with UV first (Agilent 1100 HPLC with diode array detector), followed by the LS (Wyatt Technology DAWN EOS Multi-angle laser light scattering detector) and RI (Wyatt Technology Optilab REX Differential Refractometer) detectors. For sample analysis, 150 µg of each sample was injected to the SEC-MALS system (GE Healthcare Superdex SEC column, 10 × 300 mm) in a mobile phase consisting of 25 mM Histidine, 150 mM NaCl at pH 6.8 at a flow rate of 0.5 mL/minute run at ambient temperature. Data were analyzed with Wyatt Technology ASTRA software, version 5.3, according to the two-detector method, in the case of the individual scFvs, or the three-detector method for the scFv-Fc-scFv.46
The propensity of the antibody fragments and bispecific antibodies to aggregate was evaluated using dynamic light scattering (DLS), a technique that measures the hydrodynamic size of molecules in solution as a function of diffusion. Due to the extreme sensitivity to very large species, DLS can be used to reliably quantify large aggregate down to at least 0.01% by weight in a sample containing a range of species. Each of the bsAbs was analyzed in a generic buffer solution (25 mM histidine, 125 mM NaCl, pH 6.8). No attempt was made to improve stability through formulation. Samples were stored at −80°C prior to thawing and underwent one freeze-thaw cycle. Each sample was thawed at room temperature and scanned after dilution to 1 mg/mL in the standard buffer or after spin-concentrating to 25 mg/mL. All samples were centrifuged briefly at 16,000 xg and immediately prior to DLS analysis (DynaPro Plus Plate Reader, Wyatt Technology).
All binding kinetics and affinity studies were performed on a Biacore T-100™ system, (GE Healthcare). Affinity analyses were performed by capturing the purified anti-VEGFA and anti-PDGFRβ scFvs on anti-His/Myc immobilized sensor chip. Anti-His (GTX18184, Gentex, Irvine, CA) and anti-Myc (GTX20032, Genetex, Irvine, CA) antibodies were mixed in 1:1 molar ratio and covalently immobilized to CM5 chip using amine coupling chemistry (EDC:NHS) to a density of approximately 7500RU. The anti-VEGF-A or the anti-PDGFRβ scFvs were then diluted to 10 nM and injected onto separate flow cells of the sensor chip at a flow rate of 10 uL/min for 1 minute. Serial 1:3 dilutions of VEGF-A or PDGFRβ-Fc from 33.3 nM-0.14 nM were injected over this surface and allowed to specifically bind to captured antagonists on the sensor chip. Injections of each target concentration were performed in duplicate at a flow rate of 30 uL/min with 5 minutes of association and 10 minutes of dissociation time. These experiments were performed at 25°C in a buffer of 10 mM HEPES, 500 mM NaCl, 3 mM EDTA 0.05% Surfactant P20, 1 mg/ml bovine serum albumin, pH 7.4. Between cycles, the flow cells were washed with 50 mM H3PO4 to remove the captured antagonists from the immobilized antibody surface. Resulting binding curves were globally fit to a 1:1 binding model to calculate the association and dissociation rate constants.
Affinity measurements of the tetravalent bsAb with PDGFRβ were performed with an Fc capture format. Goat anti-human IgG Fc-gamma (Jackson ImmunoResearch, West Grove, PA) was covalently immobilized on a CM4 chip (GE Healthcare/Biacore, #BR-1005-34) to a density of approximately 3500 RU using amine coupling as described above. The bsAb was captured onto a single flow cell of the CM4 chip at a density of 100 RU. Serial 1:3 dilutions of monomeric PDGFRβ (prepared from a Lys-C digest of PDGFRβ-Fc) were prepared from 100 nM-0.015 nM and injected over the surface at a flow rate of 30 µL/min. Duplicate injections of each antigen concentration were performed, with an association time of 9 minutes and dissociation time of 15 minutes. Between cycles, the flow cells were regenerated with a 30 µL injection of 20 mM HCl. The PDGFRβ binding curves were fit to a 1:1 binding model.
Affinity measurements of the bsAb with VEGF-A were performed with the target antigen covalently immobilized to the sensor chip. VEGF-A was immobilized on a CM4 chip to a density of approximately 15 RU using amine coupling as described above. Serial 1:3 dilutions of the bsAb were prepared from 100 nM-0.015 nM and injected over the surface at a flow rate of 30 µL/min. Duplicate injections of each bsAb concentration were performed, with an association time of 9 minutes and dissociation time of 15 minutes. Between cycles, the flow cells were regenerated with 3 × 30 µL injections of 10 mM glycine, pH 1.5. The VEGF-A binding curves were fit to a bivalent analyte model, and the monovalent component of the interaction (ka1, kd1) was reported.
For co-binding analysis of the bsAb, a high density (1000 RU) of VEGF-A was covalently immobilized onto a CM4 sensor chip using amine coupling as described above. The bsAb was diluted to 100 nM and injected over the immobilized VEGF-A surface at a flow rate of 10 µL/min for 5 minutes. The capture level of the bsAb was approximately 700 RU. PDGFRβ-Fc (500 nM) was then flowed over the surface for 10 minutes at a flow rate of 30 µL/min. Between cycles, the flow cells were regenerated with 3 × 30 µL injections of 10 mM glycine, pH 1.5.
All data were evaluated using Biacore Evaluation software to define either the affinity or the kinetics of the interactions. Baseline stability was assessed to ensure that the regeneration step provided a consistent binding surface throughout the sequence of injections. Binding curves were normalized by double-referencing and the resulting binding curves were globally-fit using the Biacore Evaluation Software v1.1.1.
Cytodex-3 beads (Sigma-Aldrich, St. Louis, MO) were coated with HUVEC cells (Lonza) overnight, and then embedded (200 beads/well) with human mesenchymal stem cells (Lonza, Walkersville, MD, 40 K cells/well) in fibrin gel in 12-well tissue culture plates. A 1:1 mixture of fresh EGM-2 complete media (Lonza) and D551 fibroblast conditioned EGM-2 media were added on top of the fibrin gel along with 2 ng/mL of recombinant human HGF. Medium was replaced every two days till the end of the experiment. Bevacizumab, anti-PDGFRβ E9899 or anti-PDGFRβ/VEGF-A scFv-Fc-scFv were added to the culture medium at the indicated concentrations starting from day 8 (after EC sprouts and pericyte covering were formed). Seven days after addition of antagonists, cells were fixed in 4% PFA overnight at 4°C. Endothelial cells were stained with anti-PECAM (R&D systems, BAF806), followed by fluoroscene-conjugated secondary antibody, and pericytes were labeled with anti-αSMA-Cy3 (Sigma, C6198). Cells were then viewed by an inverted fluorescence microscope and 6 × images were captured. A representative set of ten beads/well for each condition were chosen randomly. The total length of all the sprouts around a bead was measured in MetaMorph (version 220.127.116.11) by utilizing the angiogenesis tube formation application.
HUVECs were plated in 96-well flat-bottom plates (Falcon, Colorado Springs, CO) at a density of 1,000 cells per well. The HUVEC cells were plated for two days in complete EGM-2 MV media (Lonza, Walkersville, MD) at 37°C, 5% CO2. The cells were starved of serum using serum free media (SFM:DMEM-F12 (1:1) with 1x insulin-transferrin-selenium, Invitrogen, Carlsbad, CA) for 24 h and then stimulated for 24 h with 2.6 nM human VEGF-A165 with or without the serially diluted bevacizumab, anti-VEGF-A scFv, or the scFv-Fc-scFv at concentrations from 0.0005 nM to 500 nM. Cells were then pulsed with 1 µCi per well of 3H-thymidine for an additional 24 hrs at 37°C, 5% CO2. Human brain vascular pericytes (HBVPs) were seeded in 96 well flat-bottom plates (Falcon, Colorado Springs, CO) at a density of 500 cells/well in complete media (ScienCell Pericyte Media (PM) plus ScienCell supplements Fetal Bovine Serum, Pericyte Growth Supplement, and Penicillin-Streptomycin) at 37°C in 5% CO2 for 24–48 hours. The cells were starved of serum using SFM for 24 h and then stimulated with 0.4 nM human PDGF-BB in the presence or absence of a control rat anti-human PDGFRβ antibody E9899, anti-PDGFRβ scFv or the anti-PDGFRβ/VEGF-A scFv-Fc-scFv at concentrations from 2,000 nM to 0.02 nM. After 18–24 hours, 1 µCi 3H-thymidine (Amersham, Piscataway, NJ) was added to each well and cells were incubated for 3–6 hours. Cells were harvested onto filter plates and incorporation of 3H-thymidine was determined using a Packard Topcount machine. Data analysis was done using GraphPad Prism software (LaJolla, CA).
Unanesthetized female C.B-17 SCID mice were injected intravenously (i.v.) with 100 µg of anti-PDGFRβ/VEGF-A scFv-Fc-scFv in a 100 µl volume. Whole blood was collected at various times by cardiac puncture. Serum was generated and stored at −80°C. An enzyme linked immunosorbent assay (ELISA) was used to analyze test samples. The ELISA used recombinant human VEGF-A to capture scFv-Fc-scFv or appropriate standards. A biotinylated goat anti-human Fc antibody was used to bind captured Fc-containing protein, followed by incubation with streptavidin-HRP and the substrate tetramethylbenzidine. The colorimetric read-out was analyzed. The resulting concentration versus time profile was subjected to noncompartmental PK analysis using WinNonlin 5.0.1 (Pharsight Inc., Mountain View, CA). Values for the area under concentration versus time curves extrapolated to infinity (AUCINF) were calculated using the linear trapezoidal method with uniform weighting.
Groups of C.B-17 SCID mice (8–12 weeks of age) were injected subcutaneously with 2 × 106 A673 cells in the mammary fat pad. For prophylactic treatment, cohorts of mice (n = 10/gp) were injected intraperitoneally twice a week with varying concentrations of bevacizumab or anti-PDGFRβ/VEGF-A scFv-Fc-scFv in a 100 µL volume starting one day after tumor inoculation. Mice received a total of eight treatments. For the therapeutic treatment, cohorts of mice (n = 10/gp) were injected intraperitoneally twice a week with varying concentrations of bevacizumab or the scFv-Fc-scFv in a 100 µL volume, starting with a tumor volume of 150–200 mm3. Mice received a total of five doses before termination of the experiment. Weight of the mice and tumor volume were assessed three times per week. Tumor volume was calculated as (L × (W)2), where W = width and L = length of the tumor. The width was the shortest measurement of the two. Data analysis was performed using GraphPad Prism software.
HBVP were grown on glass chamber slides in PM (ScienCell Pericyte Media plus ScienCell supplements) at 37°C and 5% CO2 until they were just sub-confluent. The αPDGFRβ/αVEGF-A scFv-Fc-scFv and control mouse αhuman PDGFRβ antibody were diluted to 1 ug/mL in binding buffer consisting of DMEM + 3% BSA and Hepes buffer.
The binding of α PDGFRβ/anti-VEGF-A scFv-Fc-scFv and control αPDGF antibody was done on ice for one hour. The “time zero” (T0) slide to measure cell surface receptor expression was prepared by washing with cold PBS and fixing with 3.5% paraformaldehyde. The remaining slides were incubated at 37°C, then removed and fixed in a similar fashion at 30 min, 60 min, 1.5 hour and two hour time points. All slides were kept on ice after fixation. Once all of the slides had been fixed, they were washed one time with PBS and permeabilized for two minutes with cold methanol. The slides were washed again with cold PBS and incubated at room temperature for five minutes in 50 mM glycine, washed with PBS and blocked in 10% normal goat serum in PBS (#S-1000, Vector Labs, Inc., Burlingame, CA). Following blocking, Alexafluor 488 goat anti-mouse (Molecular Probes, Eugene, OR), or Alexafluor 488 goat anti-human (Molecular Probes, Eugene, OR) antibodies were diluted 1:150 in PBS + 0.1% Tween 20 and 0.1% BSA. Cells were labeled with appropriate antibody and incubated in the dark at room temperature for 45 minutes. Each slide was washed three times by soaking in PBS for five minutes at room temperature. Slides were mounted in Vectashield + Dapi Nuclear Stain (Vector Labs, Inc., Burlingame Calif.) and examined under the fluorescent microscope using Metavue software (Molecular Devices, Sunnyvale, CA).
Previously published online: www.landesbioscience.com/journals/mabs/article/10498