Abstract

Open in a separate window

Ferritin is a multivalent, self-assembling protein scaffold found in most human cell types, in addition to being present in invertebrates, higher plants, fungi, and bacteria, that offers an attractive alternative to polymer-based drug delivery systems (DDS). In this study, the utility of the ferritin cage as a DDS was demonstrated within the context of T cell agonism for tumor killing. Members of the tumor necrosis factor receptor superfamily (TNFRSF) are attractive targets for the development of anticancer therapeutics. These receptors are endogenously activated by trimeric ligands that occur in transmembrane or soluble forms, and oligomerization and cell-surface anchoring have been shown to be essential aspects of the targeted agonism of this receptor class. Here, we demonstrated that the ferritin cage could be easily tailored for multivalent display of anti-OX40 antibody fragments on its surface and determined that these arrays are capable of pathway activation through cell-surface clustering. Together, these results confirm the utility, versatility, and developability of ferritin as a DDS.

Introduction

Drug delivery systems (DDS) are becoming increasingly important in the advancement of novel therapeutics. DDS can be defined as systems that enhance therapeutic substances by modulating their pharmacokinetics (PK), toxicity, stability, or efficacy. DDS are typically leveraged for one or more of the following reasons: to enhance absorption into specific tissues, to prevent degradation in the biological environment, to improve intracellular delivery, or to control PK and drug distribution profiles. While synthetic polymeric scaffolds are commonly employed in DDS, their biocompatibility and the fate of the polymer metabolites are often of concern. On the other hand, there exist a myriad of natural polymer systems, including proteins, starch, cellulose, hyaluronic acid, and DNA which offer attractive alternatives.

Among them, protein-based nanocages such as viruses, ferritin, and others are of particular interest due to the cage-like structures being formed through self-assembly of protein subunits, which offers the advantage of being able to be made entirely recombinantly. Ferritin, a member of the iron storage family of proteins, is made up of 24 subunits held together by noncovalent interactions, arranged into an icosahedral cage with 2-, 3-, and 4-fold axes of symmetry around a central cavity.^1,2 Ferritin possesses many attractive features³⁻⁶ as a DDS. First, it is amenable to both chemical conjugation⁷⁻⁹ and molecular engineering techniques, for example, through the addition of a peptide or protein tag.¹⁰⁻¹² Second, ferritin is highly conserved among higher eukaryotes and therefore poses minimal risk of toxicity or immunological activity.⁷ Third, because both the monomer and the assembled cage are monodisperse entities, it is simpler to analyze compared with other delivery technologies such as polydisperse polymers or liposomes. Perhaps most importantly, the multivalency of displayed amino acid side chains, either on the external surface or internally when iron is not being stored in its central cavity, endows ferritin with the potential to carry therapeutic cargo.

Members of the tumor necrosis factor receptor superfamily (TNFRSF) have long been attractive targets for the development of anticancer therapeutics. They operate in many organ systems and play a prominent role in important biological processes including regulation of immunity, cell proliferation, and cell death.¹³ These receptors are endogenously activated by ligands that occur as trimeric transmembrane proteins or as soluble trimeric molecules, and oligomerization and cell-surface anchoring are known to be essential aspects of targeted agonism of this receptor class.¹³⁻¹⁶ Because cognate ligands of these receptors have generally made suboptimal drugs due to poor production, stability, or PK, thus far, the most common therapeutic modality advanced for this receptor class has been the monoclonal antibody (mAb). However, a significant challenge for the pharmacologic agonism of these receptors by mAbs is the general requirement for receptor oligomerization or superclustering to achieve valencies beyond two.¹⁷⁻¹⁹ While cross-linking of agonist anti-TNFRSF antibodies can be achieved in vitro by artificially coating on plates or by using secondary cross-linking reagents, in vivo cross-linking activity generally relies on extrinsic cross-linking from Fc engagement with Fc γ receptors (FcγRs) on immune cells.^16,20,21 This has had a significant impact on the clinical development of therapeutic antibody formats attempting to activate TNFRSFs. Despite the advancement of a large pipeline of TNFRSF agonist antibodies into development, none has been approved or reached late stage clinical trials, and several have been withdrawn.^{13,14,16,22−24}

While a variety of IgG-like and non-IgG-like formats have been developed,²⁵ including triabodies, tetrabodies, pentabodies, self-assembling hexameric IgG antibodies, nanoparticle and cross-linked micelle conjugates,²⁶⁻³¹ to expand valencies beyond 2, developability considerations relating to PK, safety, and physical and chemical properties have impeded advancement of these formats to the clinic. Thus, a need for multivalent delivery strategies with favorable developability properties still exists, enabling the in vivo potency needed to drive clinically relevant efficacy. Since ferritin is a biocompatible, naturally occurring icosahedron, it is an attractive scaffold for multivalent engagement of TNFRSF receptors with the potential to facilitate the superclustering necessary to drive TNFRSF signaling.

In the present study, a recombinant version of the human ferritin light chain (rFer₂₄) was designed that includes a reactive glutamine residue embedded in a peptide sequence (Q-tag) recognized by microbial transglutaminase (mTG). In parallel, a lysine residue was embedded in an mTG-reactive peptide sequence (K-tag) into the C-terminus of the heavy chain of a recombinant anti-OX40 human IgG1 antibody fragment (Fab). Site-specific conjugation of Fab-K-tag to each ferritin monomer (rFer_m) containing a Q-tag proceeded through enzymatic bioconjugation catalyzed by mTG to form a Q–K isopeptide bond, resulting in a multivalent ferritin–Fab conjugate. In-depth characterization of rFer₂₄ and rFer₂₄-Fab conjugates were performed under native and denaturing conditions. The in vitro agonist activities of rFer₂₄-Fab conjugates with varying anti-OX40 Fab densities were compared to that of a previously developed IgG format that depends on extrinsic cross-linking both in vitro and in vivo. Moderate agonism was reported for a valency of 4, while highly potent activity was observed only for a hexameric IgG format, which had a valency of 12.²⁸ The objectives of the present study were to (i) investigate whether the rFer₂₄-Fab conjugate was capable of in vitro agonism of the OX40 pathway without relying on extrinsic cross-linking and, if so, (ii) determine how activity depends on Fab valency. Positive outcomes confirm that ferritin is a promising DDS for multivalent agonism.

Results and Discussion

Results

A recombinant ferritin construct was designed with the following, starting from the N-terminus (N-term): a signal sequence for protein secretion, followed by an 8xHis tag with tobacco etch virus (TEV) protease cleavage site, followed by a microbial transglutaminase (mTG) recognition sequence (Q-tag, VLQSP)³² with a flexible linker leading into the mature N-term of liver-derived ferritin light chain (LC) (SI.1). Following expression in Expi293 cells and Ni Sepharose excel purification, the His-tag was cleaved off using TEV protease, and the digested mixture was analyzed via liquid chromatography in-line with mass spectrometry (LC/MS) under denaturing conditions (Figure ​Figure11). As shown in Table 1, theoretical and measured masses are in good agreement with one another. Moreover, the additional minor masses visible in the top panel of Figure ​Figure11A correspond to His-tag truncations (SI.2); a phenomenon that has been observed when His-tagged proteins are expressed in mammalian cells, such as Expi293 cells.

Open in a separate window

Figure 1

(A) Deconvolution from LC/MS analysis after digestion using TEV protease, with a zoomed-in view in the bottom panel. Identified masses have been assigned. (B) Size exclusion chromatography in-line with a quasi-elastic light scattering detector (SEC-QELS) of rFer₂₄ with His-tag (red trace) and after TEV protease digestion (blue trace), as well as rFer₂₄-anti-OX40 Fab conjugate (1) and unconjugated Fab (2) from reaction 1 (green trace), overlaid with measured R_H values across each peak. Averaged R_H values are also reported in the legend. (C) Physical stability profiles for rFer₂₄ and rFer₂₄-anti-OX40 Fab₂₄ were measured in phosphate-buffered saline (PBS) at 25 °C over 14 days.

Table 1

Theoretical and Measured Molar Masses for rFer_m, Anti-OX40 Fab and rFer_m-xOX40 Fab

	theoretical molar mass (Da)	measured molar mass (Da)	delta (Da)
8xHis-rFerm	23,022.74	23,023.31	0.57
rFerm	21,073.69	21,073.94	0.25
anti-OX40 Fab	47,742.28	47,742.64	0.36
rFerm-xOX40	68,815.97	68,800.93	–15.04

Open in a separate window

The recombinant ferritin cage (rFer₂₄) was also analyzed under native conditions using size exclusion chromatography in-line with a quasi-elastic light scattering detector (SEC-QELS), both before and after cleavage of the N-term His-tag (Figure ​Figure11B, red and blue traces). As shown in the UV spectra, rFer₂₄ did not exhibit aggregation, with or without the His-tag. Moreover, the SEC-QELS analysis corroborated monodispersity within each pool, as demonstrated by the narrow R_H distributions across each peak.

Conjugation to rFer₂₄ (Q-tag, VLQSP) was accomplished using an anti-OX40 Fab with mTG amine recognition sequence (RSKLG)³³ engineered into the C-terminus (C-term) of the Fab heavy chain (HC) (Figure ​Figure22A). All mTG conjugations were set up on an rFer_m basis, with initial conditions shown in Table 2 (reaction 1). SEC-QELS analysis of reaction 1 (Figure ​Figure11B, green trace) revealed a clear shift in retention time at A280 nm relative to rFer₂₄ (peak 1), as well as an excess of the Fab substrate (peak 2). The increased average R_H from 7.4 nm (rFer₂₄) and 3.3 nm (anti-OX40 Fab) to 10.9 nm also corroborated successful conjugation, while the overlaid R_Hs across peak 1 suggested that there was also some minor polydispersity within the conjugate peak. Finally, analysis of physical stability over 14 days in PBS revealed only minor changes to the main peak area, demonstrating the stability of both rFer₂₄ and rFer₂₄-anti-OX40 Fab pools at physiological pH (Figure ​Figure11C).

Open in a separate window

Figure 2

(A) Reaction scheme showing rFer₂₄ in magenta (Q-tag in red), anti-OX40 Fab (K-tag in cyan), and assembled rFer₂₄ + 24 Fabs. (B) Chromatogram of absorbance at 280 nm demonstrating resolution by reverse phase (RP) between unreacted anti-OX40 Fab (1.7 min), rFer_m-anti-OX40 Fab (2 min), rFer_m (2.3 min), and rFer_m with N-term His-tag (2.9 min) using a molar ratio of 1:0.3. (C) Representative deconvolution (top panel) with a zoomed-in view (bottom panel) around the mass of the rFer_m-anti-OX40 Fab conjugate and other minor conjugate masses.

Table 2

Reaction Conditions and Calculated Yields

	concn rFerm (μM)	concn Fab (μM)	concn mTG (nM)	stoichiometric ratio rFerm:Fab	yield
reaction 1	50	300	350	1:6	N/D
reaction 2	350	400	350	1:1.1	96.1
reaction 3	400	200	350	1:0.5	37.6
reaction 4	400	120	350	1:0.3	19.2
reaction 5	400	80	350	1:0.2	14.2

Open in a separate window

To gain a quantitative understanding of conjugate Fab density, an analytical reverse phase (RP) method was developed, yielding baseline resolution of rFer_m-Fab, Fab, and rFer_m (Figure ​Figure22B). By integrating each UV peak, this method yielded a percent Fab density based on calculating the amount of Fab-modified rFer_m versus unmodified rFer_m. With the analytical method in place, further optimization of the conjugation was accomplished by increasing the reaction concentrations of the rFer_m and Fab substrates. As shown in Table 2, adjusting substrate concentrations in reaction 2 led to nearly complete saturation of conjugation sites with a calculated Fab density of 96.1% as determined by the RP method. Conjugations with different ratios of Fab to rFer_m were also tested, and the resulting yields are summarized in Table 2 (reactions 3–5). In general, the targeted ratios and the resulting Fab densities were in agreement with only minor differences. In addition to quantitation, peak identities were confirmed by deconvolution of ion count spectra using time-of-flight mass spectrometry (TOF MS), and a mass corresponding to rFer_m-Fab was readily detected in deconvolutions of all conjugation reactions (Table 1 and Figure ​Figure22C). Secondary conjugation products, where rFer_m with a ragged N-term His-tag reacted with Fab, were also detected, suggesting that the removal of the His-tag is not necessary for the conjugation to occur. The detection of the rFer_m dimer and trimer in the deconvolutions suggest that there may also be a reactive K-site on rFer_m, which led to low levels of rFer_m cross-linking. Overall, these results demonstrated that site occupancy could be controlled by titrating the molar ratio of Fab and that conditions could be identified that achieved nearly complete site occupancy (96.1%). To further confirm site-specific conjugation, the rFer₂₄-anti-OX40 Fab₂₄ conjugate was also assessed by peptide mapping, and site-specific addition of K from the Fab to Q from rFer_m was the predominant peptide mass detected (S1.3 and Table S1).

To gain a better understanding of spatial distribution, a rFer₂₄-Fab₂₄ homology model was constructed, including 24 Fabs with site-specific isopeptide bond formation (Figure ​Figure33). Visualization of assembled rFer₂₄-Fab₂₄ gives a good sense of the spatial packing of the Fabs decorating the surface of the rFer₂₄ cage. Moreover, rFer₂₄-Fab₂₄ was also analyzed using negative stain and cryo-EM (Figures ​Figures44 and S4). Clear, high-resolution density can be seen for the rFer₂₄ cage in both negative stain and cryo-EM class averages, suggesting that the cage is intact and stable. On the other hand, as shown in Figure ​Figure44, the Fabs are blurred in the averages indicating that they are flexibly linked to the rFer₂₄, with much conformational freedom. Furthermore, using back projections of the homology model allowed for comparison of two-dimensional (2D) particle topography with the 2D averages calculated from the negative stain and cryo-EM data. Together these results give further insight into the morphology of the DDS. They also validate the suitability of this approach for multivalent conjugation by demonstrating that rFer₂₄ is amenable to site-specific addition of up to 24 Fabs and that the resulting conjugate can be developed for therapeutic applications requiring multivalency.

Open in a separate window

Figure 3

PYMOL rendering of rFer₂₄-anti-OX40 Fab₂₄ from the homology model PDB file, where the rFer₂₄ cage is in magenta and 24 repeats of the Fab moiety are in green.

Open in a separate window

Figure 4

(A) Representative 2D class averages of negatively stained rFer-OX40 and Fab₂₄ conjugates. (B) Representative 2D class averages of rFer-OX40 Fab particles were imaged using cryo-EM. (C) Back projections of the homology model are presented in Figure ​Figure33.

Finally, anti-OX40 formats were tested for their ability to promote receptor signaling using engineered Jurkat cells expressing human OX40 and an NF-κB luciferase reporter (Figure ​Figure55). While cross-linking the natural ligand with a secondary antibody (OX40L-Fc XL) resulted in potent receptor activation, monovalent anti-OX40 Fab or bivalent anti-OX40 IgG1 failed to agonize OX40. In contrast, all ratios of rFer₂₄-Fab led to receptor activation. At the highest Fab density (rFer₂₄-anti-OX40 Fab with 96.1% Fab density), agonism was comparable to that of the Fab density of the OX40L-Fc XL and did not exhibit the pronounced decrease in activity observed for OX40L-Fc XL at higher concentrations. In addition, the lower Fab:rFer₂₄ ratios demonstrated that potency is dependent on the Fab density and highlighted the ability of the rFer₂₄ cage to titrate agonism. Finally, it should be noted that rFer₂₄ without conjugated Fab did not produce any activity.

Open in a separate window

Figure 5

OX40+ Jurkat reporter assay. RLU denotes relative luminescence. Molecule concentrations (nM) were calculated as follows: rFer₂₄-Fab conjugate and aOX40 Fab concentrations were based on Fab molar mass, OX40L-Fc XL concentration was based on OX40L-Fc XL molar mass, aOX40 IgG1 concentration was based on the antibody molar mass, and rFer₂₄ concentration was based on the molar mass of the assembled ferritin cage.

Discussion

Several ferritin-based vaccines are in clinical trials,³⁴ and ferritin is also in-use for delivery of chemotherapy, immunotherapy and imaging.³⁵⁻³⁷ The work presented here extends its use cases to another therapeutic setting through the multivalent display of Fabs to induce agonism through receptor superclustering. As a recombinant protein, rFer₂₄ was amenable to protein engineering, where an amino acid sequence for site-specific conjugation was fused onto the rFer_m N-terminus. Native analysis of the rFer₂₄ substrate by SEC-QELS also confirmed a well-behaved, monodispersed protein. In addition, mTG was effective at catalyzing the formation of isopeptide bonds between Q on rFer₂₄ and K on anti-OX40 Fab (Figures ​Figures11B, ​B,2B,C,2B,C, and SI.2). While initial conditions were able to produce the rFer₂₄-Fab conjugate, further optimization of molar concentrations of each substrate led to a decrease in the Fab ratio (Table 2). This finding is likely related to the fact that mTG conjugation is K_m-driven, and thus, in addition to stoichiometry, the concentration of each substrate plays a role in conjugation efficiency. Finally, the lack of major aggregation suggests that rFer₂₄-anti-OX40 Fab is a stable, well-behaved molecule with a high potential for developability (Figure ​Figure11C).

To demonstrate that all 24 rFer_m subunits were occupied by Fab, a RP method was established to quantitatively determine the amount of Fab-modified rFer_m per cage (Figure ​Figure22B) and the results confirmed that nearly all 24 rFer monomers were conjugated to Fab, with 96.1% Fab coverage under optimal conditions (Table 2). We also determined that rFer₂₄ with various ratios of Fab displayed on the surface could be assembled by controlling the stoichiometry (Table 2). Traditional analytical techniques were sufficient to demonstrate a well-behaved conjugate with minimal secondary cross-linking (Figure ​Figure22C, SI.3. Table S1). While the LC/MS deconvolution revealed heterogeneity of the N-term of rFer_m, it also demonstrated that this heterogeneity did not interfere with reactivity since all forms of rFer₂₄ were able to react with Fab, as indicated by the resulting conjugate masses (Figure ​Figure22C).

Molecular modeling indicates that the assembly of 24 Fabs onto the rFer₂₄ scaffold is sterically possible, with enough inter-Fab distance to accommodate conformational changes (Figure ​Figure33). To build the homology model, the Fabs were manually placed near the engineered sequence for isopeptide bond creation; therefore, the current representation of the system is not necessarily energetically realistic. Interdomain interactions such as pH-dependent electrostatic and van der Waals interactions between the Fab domain could induce conformational changes limiting the accessibility of conjugation sites and could also impact the dynamics of the Fabs, potentially influencing oligomerization or superclustering. Due to the large size of the system (number of atoms >300 K), a thorough yet practical analysis of these effects requires coarse-grained modeling³⁸ or atomistic multiscale modeling using implicit solvent models,³⁹ which are beyond the current scope of our study. Nevertheless, the current model confirms the viability of spatial arrangement of the Fabs around rFer₂₄ with enough inter-Fab distances to allow conformational flexibility, as supported by the negative stain and cryo-EM data.

Indeed, the negative stain and cryo-EM micrographs allowed for further understanding of the overall morphology of the system (Figures ​Figures44 and S4) and confirmed the validity of the homology model. The stark contrast in resolution between the rFer₂₄ cage and the surrounding Fabs suggests that while the cage is in a relatively rigid conformation, the Fabs are able to move around in a variety of orientations. Moreover, back projections of the homology model allowed comparison with the 2D averages calculated from the negative stain and cryo-EM data. Although not well resolved due to their flexibility, the number of apparent Fab densities in the averages is reasonable compared to the number of Fab densities that are seen in different views of the homology model. This finding gives further insight into the conformational flexibility of the Fab domains when conjugated to the rFer₂₄ cage.

Overall, the robust activity of the conjugates demonstrates the ability of the multivalent array to take the nonfunctional Fab and turn it into one with high activity once displayed on the rFer₂₄ cage. This highlights the value of the rFer₂₄ cage as a DDS, which was amenable to addition of the TG recognition sequence and was able to accommodate the display of up to 24 Fabs on its surface. Furthermore, different ratios of rFer₂₄-anti-OX40 Fab conjugates were tested for their ability to promote receptor signaling using engineered Jurkat cells expressing human OX40 and an NF-κB luciferase reporter (Figure ​Figure55). While cross-linking the natural ligand with a secondary antibody (OX40L-Fc XL) resulted in potent receptor activation, monovalent anti-OX40 Fab or bivalent anti-OX40 IgG1 failed to agonize OX40. In contrast, all ratios of rFer₂₄-anti-OX40 Fab led to receptor activation. At the highest Fab occupancy, agonism was comparable to that of the OX40L-Fc XL. Interestingly, the hook effect that was observed with the OX40L-Fc XL at Fab concentrations greater than 20 nM was less pronounced with the rFer₂₄-anti-OX40 Fab conjugates. This decrease in activity at high OX40L-Fc XL concentrations suggests that topology may play a role in agonizing the receptor pathway. At low-to-moderate OX40L-Fc XL concentrations, it is likely that one trimeric ligand is bound to multiple receptors, whereas at high OX40L-Fc XL, it is more likely that only one receptor is bound to each trimeric ligand, leading to less potent signaling compared to the ≥2 receptors per complex case. In contrast, we hypothesize that although the anti-OX40 Fabs maintain a high degree of conformational freedom, their colocalization on the rFer₂₄ cage impacts their interaction with the receptors by enabling a single rFer₂₄-Fab array to cluster multiple receptors at higher molar concentrations than for OX40L-Fc XL. These observed differences suggest that geometry may play a role in pathway activation where colocalization of Fabs on the same scaffold may enhance clustering of OX40 receptors at saturating receptor concentrations. Moreover, the modulation of activity based on Fab percent occupancy was also demonstrated, highlighting the ability of the rFer₂₄ cage to titrate agonism. Although, it was surprising to observe that rFer₂₄-anti-OX40 (96.1% Fab) did not lead to agonism any greater than that of the cross-linked trimer, despite presenting up to 4 times as much Fab per molecule, this is attributed to reaching the upper limit of the assay range.

Overall, this study demonstrates the suitability of the rFer₂₄ DDS for the pharmacologic agonism of this important therapeutic target class. Because cognate ligands of these receptors have generally made suboptimal drugs due to poor production, stability, or PK and mAbs have failed in the clinic due to a lack of efficacy, the rFer₂₄ DDS constitutes a promising platform for multivalent delivery of Fabs. While therapeutic validation requires in vivo studies, this work corroborates previous proof-of-concept studies^27,28 and further establishes the general utility of the ferritin cage as a DDS.

Materials and Methods

RP-MS

RP-MS was used to determine the rFer_m:Fab ratios. For all gradients, mobile phase A consisted of 0.05% trifluoroacetic acid (TFA) in water, while mobile phase B consisted of 0.05% TFA and 100% acetonitrile (ACN). Resolution of rFer, rFer_m-Fab conjugates and impurities was accomplished using a reverse phase (RP) C4 column (Waters, Acquity HPLC Protein BEH C4, 2 × 150 mm), consisting of a hydrophobic alkyl chain in the stationary phase. The number of Fabs per rFer₂₄ cage was determined by quantifying the amount of unconjugated, free rFer_m using a standard curve from UV280 peak integrations of unmodified rFer_m on RP. The concentration of rFer_m-Fab was then calculated using eq 1

1

where 0.74 is the calculated extinction coefficient of rFer_m and 1.12 is the extinction coefficient for rFer_m-Fab. The percentage of rFer_m-Fab conjugates (% Fab) was determined by using eq 2.

2

In addition, mass data were acquired as described previously³² using an Agilent 1290 Infinity HPLC in tandem with an Agilent 6230 ESI/TOF mass spectrometer, operating in positive ion mode.

Acknowledgments

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.bioconjchem.4c00020.

• Plasmid sequence for liver-derived ferritin LC; full deconvoluted spectrum of rFer_m; peptide mapping of isopeptide bond between rFer_m and anti-OX40 Fab and accompanying table with detected peptides; and representative micrographs, raw particles, and class averages of rFer₂₄-anti-OX40 Fab₂₄ conjugate in negative stain and cryo-EM (PDF)

Notes

Funding for this project was provided by Genentech Inc.

Notes

The authors declare no competing financial interest.

Notes

Notes All authors are current or former employees of Genentech, a member of the Roche Group, and are shareholders in Roche. This study was supported by internal Genentech funds, and the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References