Targeted Drug Delivery with an Integrin-Binding Knottin–Fc–MMAF Conjugate Produced by Cell-Free Protein Synthesis

This article is featured in Highlights of This Issue, p. 1153

Antibody–drug conjugates (ADC) have risen to the forefront of cancer therapy in recent years due to their ability to combine the tumor-targeting selectivity of mAbs with the delivery of potent cytotoxic agents (1). This enthusiasm has been led by recent FDA approvals of ADCs including Adcetris (Seattle Genetics) and Kadcyla (Immunogen/Roche; refs. 2). Adcetris (brentuximab vedotin), which targets the cell surface protein CD30, is approved for relapsed or post–stem cell transplant Hodgkin lymphoma and systemic anaplastic large cell lymphoma (3). Kadcyla (ado-trastuzumab emtansine) targets the HER2 receptor, which is present on 20% to 30% of breast tumors (4). Numerous other ADCs, composed of a variety of tumor-targeting mAbs and cytotoxic payloads, are currently under development and are progressing forward in clinical trials (2). To further improve their therapeutic properties, development efforts have been focused on optimization of the drug warhead or the linker moiety that connects the drug to the antibody (5, 6). Less attention has been paid to exploration of the tumor-targeting agent aside from utilizing mAbs against different cell surface receptors.

We previously created EETI 2.5F (7), a unique engineered integrin-targeting agent, based on a cystine knot miniprotein derived from the Ecballium elaterium trypsin inhibitor II (EETI; ref. 8). Cystine-knot miniproteins (also known as knottins) are composed of 30 to 50 amino acid residues interconnected through a core of at least three disulfide bonds that confer high thermal, chemical, and proteolytic stability (9, 10). We showed that compared with other integrin-targeting agents, 2.5F binds with high affinity to α5β1 integrin and integrins containing an αv subunit (e.g., αvβ3 and αvβ5; ref. 7). As molecular imaging agents, radiolabeled or dye-labeled 2.5F generated robust tumor-specific signals with minimal off-target binding in mouse tumor models (11–13). Similar results were observed with a version of 2.5F fused to a murine antibody Fc domain (11), suggesting that a knottin–Fc fusion might be well suited as a “pseudo-antibody” for integrin-mediated drug delivery to tumors.

Here, we describe knottin–Fc–drug conjugates (KFDC) as an alternative to ADCs. In proof-of-concept studies, the integrin-binding peptide 2.5F was fused to a human Fc domain and tested as a vehicle for chemotherapeutic drug conjugation and targeted delivery to tumors. Compared with a small peptide, 2.5F–Fc offers the potential benefits of bivalent antigen binding, extended serum half-life due to larger size and FcRn-mediated recycling, and the ability to leverage conjugation strategies developed for ADCs. This work spans several facets of KFDC development and characterization, including: (i) production of functional knottin–Fc fusions using cell-free protein synthesis, (ii) introduction of a non-natural amino acid containing an azido functional group into the knottin–Fc fusions, (iii) site-specific conjugation of a model cytotoxic drug (MMAF) and noncleavable linker to this azido group using copper-free click chemistry, and (iv) characterization of the resulting KFDC fusions in tumor cell culture and animal models compared with control compounds.

EETI 2.5F (GCPRPRGDNPPLTCSQDSDCLAGCVCGPNGFCG) or non–integrin-binding control sequence CTRL (GCVTGRDGSPASSCSQDSDCLAGCVCGPNGFCG; ref. 7) were fused to a standard human IgG1 Fc sequence with a single amino acid substitution that was replaced with an amber codon at the F404 position. The sequences were codon-optimized for production in E. coli (DNA 2.0, Inc.) and cloned into a plasmid to be used in a cell-free protein synthesis reaction. U87MG and MB-468 cell lines were obtained from ATCC, and A2780 cells were obtained from Sigma-Aldrich. All cell lines were obtained from cell banks in 2015 (except A2780 cells, obtained in 2014), within 6 months of experimentation. Cell lines obtained from ATCC were tested by the manufacturer for sterility (aerobic and anaerobic cultures), human pathogens (HIV, HepB, HPV, EBV, mycoplasma, and CMV) and authenticity [cytochrome c oxidase I (COI) and short tandem repeat (STR) analysis] as per manufacturer's certificate of analysis. A2780 cells, obtained from Sigma-Aldrich, were tested by the Culture Collections division of Public Health England for sterility (aerobic and anaerobic cultures), human pathogens (mycoplasma), and authenticity (mitochondrial DNA sequencing and STR analysis). All media components were purchased from Life Technologies except where indicated. U87MG cells were grown in DMEM (cat# 11995-073) with 10% FBS (cat# 26140079) and 1% penicillin/streptomycin (cat# 15140-122). MB-468 cells were grown in DMEM and Ham F-12 medium (DMEM/F12; Gibco/Thermo Fisher; cat# 11320-033) with 10% FBS and 1% penicillin/streptomycin. A2780 cells were grown in RPMI1640 + l-glutamine (cat# 11875-093), with 10% FBS and 1% penicillin/streptomycin.

Knottin–Fc fusion constructs were produced as described previously using cell-free protein synthesis (14) with modifications detailed below. E. coli strain SBDG224 was grown in a continuous fermentation batch and converted into extract. The extract was treated with 75 μmol/L iodoacetamide for 30 minutes at room temperature (20°C). Final concentrations of components in the reaction mixture were 30% extract, 2 mmol/L para-azidomethyl-l-phenylalanine (pAMF, a non-natural amino acid), 0.5 μmol/L pAMFRS (an aminoacyl tRNA synthetase), 2 mmol/L GSSG, 2 mmol/L amino acids (except 0.5 mmol/L for tyrosine and phenylalanine), and 10 μg/mL plasmid. The reaction was incubated in a bioreactor for 16 to 18 hours at 30°C with 30% air saturation, 0.2 vvm sparged air flow, and no pH control. The protein purification process consisted of three major steps: clarification, chromatography, and formulation. In the clarification step, the harvest was centrifuged and filtered through a 0.45-μm disposable filter unit. Then, the desired protein was captured using GE Healthcare's MabSelect SuRe resin, which functions like Protein A medium. A step gradient with 0.1 mol/L citric acid at pH 3.5 was used to elute the desired protein. Subsequently, the eluate was concentrated and dialyzed into HyClone 1× PBS (Fisher Scientific). SDS-PAGE on purified proteins was carried out on NuPAGE Novex 4%–12% Bis–Tris gels (Thermo Fisher Scientific) with 1× MES buffer. To reduce samples, a final concentration of 1 mmol/L dithiothreitol (DTT) was used. Size exclusion chromatography (SEC) was performed on a Sepax Zenix-C SEC-300 column (Sepax Technologies). The SEC buffer was 100 mmol/L potassium phosphate, 150 mmol/L sodium chloride, pH 6.5. Protein synthesis yields were approximately 300 mg/L from the cell-free reaction, with about 100 mg/L recovery after purification. Note that all Fc fusion proteins used in this work contain the pAMF non-natural amino acid at position 404, based on Eu numbering of the Fc domain (15).

Purified knottin–Fc variants containing pAMF were conjugated to the cytotoxic agent MMAF through copper-free click chemistry, using a constrained cyclooctyne reagent. In brief, dibenzocyclooctyl (DBCO)-PEG₄-MMAF (ACME Bioscience Fig. 1D) was dissolved in DMSO to a final concentration of 5 mmol/L. The compound was added to the purified protein sample in PBS buffer at a drug to protein molar ratio of 6 to 1. The mixture was incubated at room temperature (20°C) for 17 hours. Excess free drug was removed by Zeba plates (Thermo Fisher Scientific) equilibrated in PBS.

To determine drug loading, 1 mg/mL of knottin–Fc–MMAF in PBS was diluted 1:4 (v:v) into 7.2 mol/L guanidine-HCl, 0.3 mol/L sodium acetate, pH 5.3, and reduced by adding TCEP (Thermo Scientific) at final concentration of 10 mmol/L. The mixture was incubated at 37°C for 15 minutes in an Eppendorf Thermomix R while shaking at 300 rpm. A total of 2 μg of sample was injected into an Agilent ZORBAX C8 column (1.8 μm, 300 Å, 100 × 2.1 mm² from Agilent), connected to an U-HPLC system containing a binary gradient pump, temperature-controlled column compartment, autosampler, and a diode array detector. The RP-UHPLC assay was performed at 0.5 mL/minute at 80°C using 0.1% trifluoroacetic acid (TFA) in water (mobile phase A, MPA) and 0.1% TFA in acetonitrile (ACN; mobile phase B, MPB), and monitored at absorbance of 280 nm. The 13-minute method consists of a 3-minute isocratic hold at 30% MPB, a linear gradient with a 4%/min increment from 30% to 50%, a 1-minute wash using 95% MPB, and a 3-minute re-equilibration at 30% MPB. Shoulders in the HPLC peaks are minor impurities or conformations. Drug–protein ratio (DPR) was determined using the following equation and was found to be about 1.9 for 2.5F–Fc–MMAF and CTRL–Fc–MMAF:

Cells were detached from culture plates using cell dissociation buffer [PBS (1.0581 mmol/L KH₂PO₄, 154.0041 mmol/L NaCl, and 5.6002 mmol/L Na₂HPO₄) containing 0.05% EDTA, Invitrogen]. Detached cells were subsequently washed with PBS. A total of 4 × 10⁴ cells per staining reaction were resuspended in 50 μL of PBS/BSA (0.1% BSA in PBS) containing a 1:100 dilution of AlexaFluor-488 anti-α_Vβ₃ integrin (R&D Systems), 1:100 dilution of AlexaFluor-488 anti-α_Vβ₅ integrin (Millipore), 1:25 dilution of FITC anti-α₅ integrin (BioLegend), or 1:200 dilution of goat anti-mouse IgG-FITC (Life Technologies) as a control. Cells were incubated on ice with primary antibodies for 75 minutes. Cells were pelleted and then washed with 1 mL of PBS/BSA. To enhance the signal and increase the dynamic range of the assay, staining reactions containing anti-α_Vβ₅ integrin were then incubated on ice with a 1:200 dilution goat anti-mouse IgG-FITC secondary antibody for 75 minutes and washed again with PBS/BSA. All samples were analyzed using flow cytometry (Guava EasyCyte 8HT flow cytometer, EMD Millipore), and the resulting data were evaluated using FlowJo software (TreeStar, Inc.). Relative fluorescence for each sample was measured as a ratio of sample signal/mean secondary only control signal. Error bars represent the SD of experiments performed in triplicate.

To quantify the integrin expression in terms of receptors/cell, the Quantum Simply Cellular (QSC) Kit anti-Mouse IgG (Bangs Laboratories, Inc.) was used. Beads were incubated with the same concentration of antibody as used on each cell line for 30 minutes at room temperature. For α_Vβ₅ integrin analysis, beads were washed with 1 mL PBS/BSA following primary incubation, then incubated for 75 minutes at room temperature in secondary antibody. Beads were then pelleted and washed with 1 mL PBS/BSA 3 times. In all experiments, beads were analyzed on the same day as the cells, using identical flow cytometer settings. Software provided with the kit was used to create a regression between fluorescence signal and antibodies/cell that enabled quantification of the fluorescence signal from each of the cell lines as number of receptors/cell (Supplementary Fig. S1). As a secondary antibody was used to boost signal for α_Vβ₅ integrin, the reported receptors/cell for that integrin are semiquantitative, as 1:1 binding of secondary to primary antibody is not definitive.

To measure the relative binding affinities of knottin–Fc fusion proteins, competition binding assays were performed as described previously with some modifications (16). AlexaFluor 488–conjugated EETI 2.5F (AF488-2.5F) was used as a competitor to compare the relative binding of 2.5F–Fc, 2.5F–Fc–MMAF, and CTRL–Fc–MMAF. Briefly, 4 × 10⁴ cells were detached with cell dissociation buffer, washed with IBB (25 mmol/L TRIS, pH 7.4, 150 mmol/L NaCl, 2 mmol/L CaCl₂, 1 mmol/L MgCl₂, 1 mmol/L MnCl₂, and 0.1% BSA) and incubated with 2 nmol/L AF488–2.5F and varying concentrations of knottin–Fc fusions in 1 mL of IBB at 4°C for 2 hours. The cell-bound fluorescence remaining after washing twice with 1 mL of PBS/BSA was determined by flow cytometry, as described above. Adjusted fluorescence values were calculated as the geometric mean of the fluorescence signal for the negative control (cells only) subtracted from the geometric mean of fluorescence signal of the sample. From this, the percentage bound was calculated as the adjusted fluorescence of each sample divided by the adjusted fluorescence of the positive control for each cell line × 100. Half-maximal inhibitory concentration (IC₅₀) values were determined by nonlinear regression analysis using Prism (GraphPad Software, Inc). Error bars represent the SD of experiments performed in triplicate.

Cells were seeded in a 96-well plate at a density of 2,000 cells per well and grown overnight at 37°C, 5% CO₂ in the media described for each cell line above. Cells were subsequently treated with 100 μL of fresh media, containing varying concentrations of knottin–Fc fusion proteins or linker-modified MMAF, and incubated for 5 days at 37°C, 5% CO₂. Cell proliferation was measured using the Cell Counting Kit-8 (CCK-8; Dojindo), by adding the water-soluble tetrazolium salt, WST-8, to each well in an amount equal to 10% of the culture volume. After incubation for 1 hour at 37°C, absorbance at 450 nm was measured with a Synergy H4 microtiter plate reader (BioTek Instruments). Cell proliferation was expressed as a percentage of absorbance relative to the control of untreated cells. We processed the data for these experiments by first subtracting a background value from each well based on absorbance of CCK-8 + media (no cells). Percent maximum proliferation was then reported as (sample − background)/(control − background) × 100. Error bars represent the SD of experiments performed in triplicate.

Animal procedures were carried out according to a protocol approved by the Stanford University Administrative Panels on Laboratory Animal Care (APLAC #22942). For tumor cell implantation, 6-week-old female nu/nu mice (Charles River Laboratory) were anesthetized with 2.5% isoflurane by inhalation with a flow rate of 1 L/minute. A volume of 100 μL of 50/50 PBS/Matrigel (Corning #356231), containing 5 × 10⁶ U87MG cells, was injected subcutaneously into the flank. Tumors were allowed to grow for 6 days until reaching a size of approximately 30 to 50 mm² in tumor area. At day 6 after inoculation, all mice were weighed and tumors measured. Mice were binned into experimental groups to ensure equivalent average tumor size and average weight across each group. All compounds tested were administered via intraperitoneal injection in 100 μL PBS, with dosing frequency and concentration dependent on therapeutic group as indicated. Tumors were measured three times weekly using digital calipers, and animal weight was recorded on each dosing day to monitor mice for weight loss as a measure of compound toxicity. Tumor area was calculated using area = x × y, where x is the longest axis of the tumor and y is the axis perpendicular. Euthanasia criteria were defined as 20% body weight loss or tumor size greater than 100 mm² in area.

Survival results were plotted using the Kaplan–Meier method. A log-rank (Mantel–Cox) test was conducted to compare the survival curve of each treatment group to that of the untreated control. The resulting P values were adjusted by the Bonferroni method to control the false-positive rate. Tumor size across groups on day 15 were analyzed using a one-way ANOVA. Each treatment group was compared with the untreated control using the Dunnett method. Statistical analyses were performed using GraphPad Prism.

EETI 2.5F was genetically fused to a human IgG1 Fc domain to create a construct termed 2.5F–Fc, which was produced by cell-free protein synthesis using an E. coli–based extract (14). A non-natural amino acid was selectively incorporated into the Fc domain by replacing position F404 with an amber stop codon that allowed incorporation of pAMF using a modified aminoacyl tRNA synthetase (ref. 14; Fig. 1A). 2.5F–Fc proteins were expressed at high levels using cell-free protein synthesis (typical reaction yields ∼300 mg/mL), and following purification were analyzed by SDS-PAGE and gel filtration chromatography (Fig. 1B and C).

To create a KFDC, the pAMF azido functional group in the Fc domain was reacted with a MMAF derivative containing a dibenzocyclooctyl (DBCO) group using strain-promoted cycloaddition copper-free click chemistry (Fig. 1D). Copper-free click chemistry was chosen to minimize oxidation of the protein by reactive oxygen species and to prevent potential cytotoxicity from residual copper in the sample. A flexible PEG linker was used to bridge the MMAF molecule and the 2.5F–Fc fusion. This noncleavable linker minimizes off-target bystander effects by delivering intact protein–drug conjugate to cells for receptor-mediated internalization and processing and thus allowed us to demonstrate the ability of the integrin-targeting 2.5F–Fc to deliver MMAF to tumor cells without potential complications of extracellular drug cleavage. Reversed-phase HPLC (RP-HPLC) analysis confirmed a DPR of approximately 2, as expected on the basis of the incorporation of one unnatural amino acid per Fc chain (Fig. 1E). A knottin–Fc–MMAF conjugate containing a scrambled integrin-binding sequence (refs. 7, 13; CTRL–Fc–MMAF) was prepared as a negative control and also possessed a DPR of approximately 2.

Three biologically diverse tumor cell lines, glioblastoma (U87MG), breast (MB-468), and ovarian (A2780) carcinomas, were tested for the presence of cell surface integrin receptors αvβ3, αvβ5, and α5β1 by antibody staining and detection using flow cytometry. High levels of αvβ3 integrin were present on U87MG cells, whereas MB-468 and A2780 cell lines expressed only minimal levels of this integrin (Fig. 2A). In comparison, high levels of αvβ5 integrin were found on MB-468 cells, with lower levels of this integrin present on both U87MG and A2780 cells (Fig. 2B). U87MG cells expressed the highest levels of the α5 integrin subunit, followed by lower levels of expression on A2780 cells and minimal expression on MB-468 cells (Fig. 2C). The α5 integrin subunit is only known to pair with the β1 subunit (17), thus measurement of α5 is indicative of α5β1 heterodimer expression. The variable expression of integrin receptors observed across these cell lines in our studies and others (18, 19) highlights a potential benefit of the broad integrin-binding specificity of EETI 2.5F for tumor-targeting applications.

Competition binding assays were performed to compare the relative binding affinities of 2.5F–Fc, 2.5F–Fc–MMAF, and CTRL–Fc–MMAF to U87MG cells. Cells were incubated with varying concentrations of knottin–Fc proteins and a constant amount of AF488-labeled 2.5F peptide competitor at 4°C to prevent internalization. Fluorescent binding signals were measured using flow cytometry and half-maximal inhibitory concentration (IC₅₀) values were determined by nonlinear regression analysis. 2.5F–Fc and 2.5F–Fc–MMAF had similar IC₅₀ values (6.9 ± 1.1 vs. 8.3 ± 1.3 nmol/L, respectively), indicating that MMAF conjugation has negligible impact on integrin-binding affinity (Fig. 3A). CTRL–Fc–MMAF did not compete AF488–2.5F binding to U87MG cells at concentrations up to 200 nmol/L, demonstrating lack of measurable integrin binding from this negative control protein (Supplementary Fig. S2). The relative binding affinity of 2.5F–Fc–MMAF to A2780 and MB-468 cell lines was also measured, with IC₅₀ values of 1.1 ± 1.2 and 1.2 ± 1.2 nmol/L, respectively (Fig. 3B and C).

Proliferation of U87MG, A2780, and MB-468 cell lines was assayed after treatment with 2.5F–Fc–MMAF and compared with the effects of 2.5F–Fc or linker-modified MMAF added separately or in combination (Fig. 4). In all cell lines tested, 2.5F–Fc treatment alone exhibits moderate inhibition of cell proliferation at the highest concentrations tested (up to 1 μmol/L), as measured by cell dehydrogenase activity, which produces a formazan dye that can be detected by spectrophotometry. The cytotoxicity of 2.5F–Fc at high concentrations, particularly in A2780 cells, and to a lesser degree MB-468 cells, is likely due to the ability of integrin-targeting molecules to disrupt cell adhesion. Similar results were seen when tumor cells were treated with high concentrations of 2.5F peptide in a previous study, in contrast with U87MG cells which are more resistant to detachment-induced apoptosis (16). Linker-modified MMAF treatment alone inhibits cell proliferation to some degree; a similar level of inhibition was observed upon coadministration of 2.5F–Fc plus linker-modified MMAF, indicating the lack of synergistic effects from combination treatment. Similar trends were observed across all three cell lines tested. The highest level of inhibition of cell proliferation occurred upon treatment with 2.5F–Fc-MMAF, with the greatest effects observed in U87MG cells, potentially due to the high integrin expression levels on these cells compared with A2780 and MB-468 cells. IC₅₀ values for 2.5F–Fc–MMAF treatment were: U87MG = 9.2 ± 1.1 nmol/L, A2780 = 26.1 ± 1.1 nmol/L, and MB-468 = 54.1 ± 1.0 nmol/L. Cells treated with CTRL–Fc–MMAF exhibit less inhibition of proliferation compared with 2.5F–Fc–MMAF or linker-modified MMAF alone (Supplementary Fig. S3). These results demonstrate integrin-targeting specificity of drug delivery mediated by 2.5F–Fc–MMAF and suggest that conjugation of MMAF to CTRL-Fc, which does not bind integrins, reduces nonspecific activity of free drug.

Next, 2.5F–Fc–MMAF was tested for its ability to inhibit tumor growth in U87MG xenograft models. U87MG cells were chosen for these studies, as they showed the greatest response to inhibition of cell proliferation upon treatment with 2.5F–Fc–MMAF. In a preliminary study, nu/nu mice (n = 5 per treatment group) were inoculated with U87MG cells in their right flank and tumors were allowed to establish for 6 days, with an average size of 35 mm² before initiation of therapy. We first evaluated different dosing amounts and schedules of 2.5F–Fc–MMAF administered via intraperitoneal (i.p.) injection: 10 or 5 mg/kg administered twice or three times per week or 1 mg/kg administered three times per week, for a period of 3 weeks. A dose-responsive effect on tumor regression was seen, with the greatest effects observed in the 10 mg/kg treatment groups (Supplementary Fig. S4). The 10 mg/kg dosing groups both exhibited 80% survival at day 50, as compared with 0% in the control mice (P = 0.0189). No significant difference in tumor regression (Supplementary Fig. S4A) or survival benefit (Supplementary Fig. S4B) was seen between two or three times a week dosing at either 5 mg/kg (P > 0.99) or 10 mg/kg (P > 0.99).

On the basis of these initial results, we performed a more extensive study, testing the in vivo efficacy of 2.5F–Fc–MMAF compared with its individual components (2.5F–Fc or linker-modified MMAF), and the nonbinding CTRL-Fc-MMAF control, all administered at equal dosing frequency and molar equivalent dosing. As above, nu/nu mice were inoculated with U87MG cells in their flank and allowed to establish for 6 days, this time with an average tumor size of 47.5 mm² before initiation of therapy. Mice were assigned to one of five groups (n = 8–9): (i) untreated control, (ii) 2.5F–Fc–MMAF, (iii) CTRL–Fc–MMAF, (iv) 2.5F–Fc (all administered at 10 mg/kg), or (v) linker-modified MMAF (0.24 mg/kg; corresponding to a molar equivalent of MMAF). Compounds were administered via intraperitoneal injection twice per week for a total of 3 weeks of dosing. The treatment appeared to be well tolerated, with no significant weight loss or adverse events observed in any of the treatment groups (Supplementary Fig. S5). Significant tumor regression was again seen with 2.5F–Fc–MMAF treatment; however, we hypothesize that the larger average tumor size at the start of dosing (47.5 vs. 35 mm²) may have affected the durability of the response compared with the pilot experiment. Day 15 was chosen for statistical comparison, as it was the final day of the study with all animals included. When comparing tumor size in treated groups compared with the untreated control at day 15 after tumor inoculation, only 2.5F–Fc–MMAF administration yielded a significant regression in growth (P < 0.0001). In contrast, no significant effect on tumor regression or survival benefit was seen in the groups dosed with 2.5F–Fc or MMAF alone, with results similar to the untreated controls (Fig. 5). The effect of CTRL–Fc–MMAF on tumor size at day 15 was not significant (P = 0.7044). A moderate but not statistically significant effect (P = 0.0772) on survival was observed in the group dosed with CTRL–Fc–MMAF compared with untreated controls. It is possible that prolonged serum half-life or Fc-mediated cell trafficking of MMAF could result from its chemical conjugation to the CTRL–Fc fusion protein. Similar moderate effects have been observed with MMAF-conjugated control antibodies in other studies (4, 20).

Members of the integrin receptor subfamily have been found to be present at high levels on a variety of tumors and tumor-associated vasculature (21, 22). These receptors, which include integrins that contain an αv subunit (such as αvβ3 and αvβ5) and α5β1 integrin, play a key role in tumor invasion, metastasis, angiogenesis, and survival, highlighting them as promising targets for therapeutic intervention (21, 23–28). Integrin antagonists targeting αvβ3, αvβ3/αvβ5, or α5β1, in the form of peptidomimetics, small molecules, or mAbs, have been the subject of intense oncology drug development efforts for decades (29–35), although none are yet FDA approved. These integrin-targeting agents were developed with the goal of inhibiting angiogenesis or blocking cell signaling, adhesion, or migration (21) but have failed to perform well in clinical trials (32, 34). A promising alternative is to instead use integrins for targeted delivery of cytotoxins to tumors.

EETI 2.5F is unique in its ability to target a broad spectrum of tumor-associated integrins, a feature that has not been achieved with any antibody to date, which instead only bind αv-containing integrins or α5β1 integrin. Targeting multiple integrin receptors is potentially beneficial for reducing drug resistance and tumor growth, as cancer cells can change their integrin repertoire in response to drug treatment (36). Integrins, particularly α6β4, α6β1, αvβ5, α2β1, and α3β1 (21), are found to be present on epithelial tissues and are also upregulated in wound healing. Our previous studies have shown that 2.5F and 2.5F–Fc, which bind with high affinity to both mouse and human integrins, exhibit high selectivity for tumor versus healthy tissue in contrast to other integrin-targeting agents (11, 13). It is possible that multispecific receptor targeting and the low level of αvβ3 and α5β1 integrins present on adult epithelial tissue compared with tumors imparts selectivity for diseased tissue, a concept that has been described in the literature for multispecific proteins (37).

First-generation ADCs were produced by randomly conjugating drugs to antibodies through sulfhydryl or amino side chains of cysteine or lysine residues (38). As examples, Kadcyla and Adcetris are produced by drug conjugation to surface-exposed lysines or partial disulfide reduction and conjugation to free cysteines, respectively (39, 40). These conjugation approaches lead to heterogeneous ADC products with varied numbers of drugs conjugated across different locations within the antibody. Heterogeneous ADCs have been shown in some cases to possess suboptimal activity and pharmacokinetic properties compared with homogeneous ADC molecules (41–43). Site-specific drug conjugation to create homogeneous ADCs has been carried out through introduction of non-natural amino acids and functional groups that facilitate orthogonal chemistries (44–46). Such chemical handles have been incorporated into recombinantly produced antibodies using cell-based (44, 47) or cell-free (14) methods, introduced cysteine residues (41), lysine analog incorporation (48, 49), or enzymatic conjugation using formylglycine converting enzyme or trans-glutaminase (39). Here, we demonstrate that functional knottin-Fc fusions can be produced by cell-free protein synthesis, followed by subsequent site-specific chemical attachment of a linker-warhead through an introduced non-natural amino acid, to produce homogeneous drug conjugates.

We found that 2.5F–Fc–MMAF exhibits lower potency in vivo in comparison with other recently reported ADCs (50, 51). This level of activity is consistent with the modest potency observed in our in vitro studies, with IC₅₀ values in the range of 10 to 50 nmol/L (Fig. 4). Lower potency could result from a number of factors including inefficient internalization that limits the amount of drug delivered inside of the cell, suboptimal trafficking that hinders localization and release of the conjugate within the lysosomal compartments, or perhaps intrinsically low potency of the large pAMF–DBCO–PEG₄–MMAF moiety that is likely the active drug compound released. Thus, while 2.5F–Fc–MMAF provides proof-of-concept data validating tumor targeting and drug delivery using a knottin–drug conjugate, further optimization will be required for development of viable clinical candidates based on this strategy. Integrin-targeted ADCs were previously evaluated with CNTO 95, an anti-αv integrin antibody (52), conjugated to maytansinoids (53). CNTO 95–drug conjugates (loading of ∼3.5 drugs per antibody) showed potent antitumor effects in lung and colon xenograft tumor models. This results of this study are difficult to directly compare with 2.5F–Fc–MMAF since drug loading, linkers, warhead, and tumor models are different; however, provide further evidence that integrins are promising targets for drug delivery.

The three components of an ADC (mAb, linker, and drug) together play a concerted role on the efficacy of the overall conjugate; thus optimization of each component has been the subject of many development efforts. In our studies, attachment of MMAF with a noncleavable linker was chosen to validate the concept of a knottin–Fc–drug conjugate as: (i) MMAF has been used in numerous previous studies describing novel therapeutic molecules (50, 51), and (ii) MMAF does not need to be cleaved from its linker to be active, allowing us to perform initial studies without complications of premature linker cleavage. An important next goal is to extend this approach to improve potency by testing parameters including alternative warheads and cleavable linkers. ADCs under clinical development mostly utilize three groups of cytotoxins including auristatins, maytansines, and calicheamicins (2), with other potent small molecules gaining traction, such as pyrrolobenzodiazepine dimers (54). In addition, most ADCs in the clinical pipeline contain linkers that are processed intracellularly though disulfide bond exchange or protease cleavage following internalization (55). Cleavable linkers do not require processing in the lysosome for activity and can help promote bystander activity through efflux of hydrophobic drugs, which can be useful for increased efficacy against solid tumors that exhibit heterogeneous target expression. Finally, the number and location of attached cytotoxins (43, 56), as well as the choice of target and internalization rate (2, 5), have also been critical features underlying ADC efficacy that can be optimized in next-generation knottin–Fc–drug conjugates.

Current efforts to advance ADC technologies have focused on novel drug conjugation methods, linker optimization, alternate warheads, or varying drug loading ratios. Less work has been carried out on the tumor-targeting agent itself beyond antibodies. Several peptide–drug conjugates are in development, such as GNR1005, a 19 amino acid peptide called Angiopep2 conjugated to paclitaxel (57), and Zoptarelin doxorubicin, (AEZS-108), a small peptide agonist of the luteinizing hormone-releasing hormone (LHRH) receptor linked to doxorubicin (ref. 58; others reviewed in ref. 59). Recent studies suggest that antibody targeting is limited by poor and heterogeneous tumor targeting driven by their large size (60). The 2.5F–Fc–MMAF protein is approximately 60 kDa in size, about 60% smaller than a traditional ADC, and thus presents a new tool to explore the effects of protein size on tumor penetration in future studies (61). Our work also highlights the potential for development of other tumor targeting drug conjugates based on so-called “alternative scaffolds,” including, but not limited to anticalins, affibodies, fibronectin domains, and designed ankyrin repeat proteins (62, 63). In this way, combining the array of available tumor-targeting proteins with emerging linker, warhead, and conjugation technologies provides numerous opportunities for the development of next-generation cancer therapeutics.

N.V. Currier is Associate Medical Director at Biogen. J.R. Cochran has ownership interest (including patents) in Stanford University. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N.V. Currier, S.E. Ackerman, J.R. Kintzing, A. Steiner, A.K. Sato, J.R. Cochran

Development of methodology: N.V. Currier, S.E. Ackerman, J.R. Kintzing, A. Steiner, A.K. Sato

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.V. Currier, S.E. Ackerman, J.R. Kintzing, R. Chen, M.F. Interrante, A.K. Sato

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.V. Currier, S.E. Ackerman, J.R. Kintzing, M.F. Interrante, A. Steiner, J.R. Cochran

Writing, review, and/or revision of the manuscript: N.V. Currier, S.E. Ackerman, J.R. Kintzing, R. Chen, M.F. Interrante, A. Steiner, A.K. Sato, J.R. Cochran

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.V. Currier, J.R. Kintzing

Study supervision: J.R. Cochran

The authors thank Wherly Hoffman for assistance with statistical analyses and Sutro team members Nina Carlos, Xiaofan Li, Cuong Tran, Sammie Lai, Gang Yin, Jeff Hanson, and Yiren Xu for assistance with protein production and characterization.

This work was funded in part by the Stanford Bio-X Interdisciplinary Initiatives Program, and seed funding from the Stanford Cancer Institute (all authors). The authors wish to acknowledge the following fellowship support: Anne T. and Robert M. Bass Endowed Fellowship in Pediatric Cancer and Blood Diseases (to N.V. Currier), Stanford Bio-X Bowes Fellowship (to S.E. Ackerman), NSF Graduate Fellowship (to J.R. Kintzing), and the Stanford Bioengineering REU program (to M.F. Interrante).

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.