Brain delivery of therapeutic proteins using an Fc fragment blood-brain barrier transport vehicle in mice and monkeys

See allHide authors and affiliations

Science Translational Medicine  27 May 2020:
Vol. 12, Issue 545, eaay1359
DOI: 10.1126/scitranslmed.aay1359

Transport vehicle for CNS therapeutics

Delivering biotherapeutics to the brain is complicated by the presence of the blood-brain barrier (BBB). Kariolis et al. and Ullman et al. developed a transport vehicle (TV) consisting of an Fc fragment engineered to bind to the transferrin receptor, a protein highly expressed at the BBB. Utilizing the TV in the form of an antibody directed against beta-secretase enhanced brain delivery and effects in both mice and monkeys following systemic administration. Systemic delivery of a TV fusion with the iduronate 2-sulfatase enzyme was effective at reducing peripheral and central pathologies in a mouse model of mucopolysaccharidosis type II. The TV platform approach potentially offers a treatment for neurological disorders.


Effective delivery of protein therapeutics to the central nervous system (CNS) has been greatly restricted by the blood-brain barrier (BBB). We describe the development of a BBB transport vehicle (TV) comprising an engineered Fc fragment that exploits receptor-mediated transcytosis for CNS delivery of biotherapeutics by binding a highly expressed brain endothelial cell target. TVs were engineered using directed evolution to bind the apical domain of the human transferrin receptor (hTfR) without the use of amino acid insertions, deletions, or unnatural appendages. A crystal structure of the TV-TfR complex revealed the TV binding site to be away from transferrin and FcRn binding sites, which was further confirmed experimentally in vitro and in vivo. Recombinant expression of TVs fused to anti–β-secretase (BACE1) Fabs yielded antibody transport vehicle (ATV) molecules with native immunoglobulin G (IgG) structure and stability. Peripheral administration of anti-BACE1 ATVs to hTfR-engineered mice and cynomolgus monkeys resulted in substantially improved CNS uptake and sustained pharmacodynamic responses. The TV platform readily accommodates numerous additional configurations, including bispecific antibodies and protein fusions, yielding a highly modular CNS delivery platform.


The central nervous system (CNS) is among the most highly perfused and vascularized organ systems in the body. The cerebrovasculature and associated brain cells comprise a neurovascular unit that couples neural activity to blood flow and provides for the development and maintenance of the blood-brain barrier (BBB) (1), an exquisitely regulated interface between the bloodstream and neural tissue. The BBB is formed, in part, by endothelial cell tight and adherens junction proteins, which severely restricts transport of most macromolecules circulating within the blood compartment (26). This limited exchange represents a key challenge for delivering peripherally administered protein therapeutics into the CNS (7). Multiple groups have observed that CNS exposure for circulating biologics is limited to 0.1 to 0.4% of corresponding serum concentrations (810), although some estimates for the transport of immunoglobulins into neural tissue are far lower (11). Consequently, the maximal brain concentration for peripherally dosed large molecules is often insufficient to achieve the target engagement required for a therapeutic response (12, 13). Administration methods such as intracranial, intrathecal, or intraventricular injections, along with chemical BBB disruption, may facilitate higher brain concentrations for some drugs, but these approaches are invasive and incompatible with repeat dosing regimens (1416). Although human genetics combined with rodent disease models has improved our ability to select promising biological targets for many drug candidates, insufficient brain exposure remains a likely cause for the failure of many antibodies and other large molecule drugs directed toward CNS diseases over the past decade (17).

Much progress has been made in developing receptor-mediated transcytosis (RMT) as a noninvasive strategy for enhancing brain exposure of biotherapeutics (12, 18). RMT is an endogenous process wherein biomolecules such as iron-bound transferrin, insulin, and lipoparticles bind to cognate receptors on brain endothelial cells and are subsequently transported across the BBB. Protein therapeutics engineered to bind these brain endothelial cell–enriched receptors can similarly exploit RMT as a means of CNS delivery. Previous RMT-based delivery strategies have relied upon antibody variable domain binding to engage brain endothelial cell receptors (12, 1820). Most of these approaches use one of two architectures: a bispecific antibody where one Fab binds the RMT receptor and the other binds a therapeutic target (21), or an antibody where both Fabs bind a therapeutic target and an RMT-binding domain is appended to a heavy or light chain terminus via a peptide linker (2225). These approaches respectively preclude bivalent and bispecific therapeutic targeting or require the appendage of an unnatural fragment to the immunoglobulin G (IgG) scaffold.

Here, we describe the development of a platform technology that we refer to as a BBB transport vehicle (TV), a human IgG1 Fc engineered to bind to RMT targets expressed on the surface of brain endothelial cells. Incorporating BBB receptor binding into the Fc domain, rather than relying on antibody variable domain binding, enables TVs to retain the natural biophysical properties and desirable pharmacokinetics (PK) of IgGs. This simultaneously allows for highly modular combinations with Fabs, yielding an antibody transport vehicle (ATV), or with other proteins, yielding a protein transport vehicle (PTV; Fig. 1A). Previous efforts to introduce completely new molecular recognition properties into the Fc domain have focused on oncology targets (26). In these previous studies, structural loops in the lower CH3 domain were engineered to generate “Fcabs,” Fc domains with antigen binding (27). High-affinity Fcabs targeting αvβ3 integrin (28), vascular endothelial growth factor (VEGF) (29), human epidermal growth factor receptor 2 (HER2) (3032), and lymphocyte activation gene 3 (LAG3) (33) have been reported, and a therapeutic containing a LAG3-binding Fcab has entered a phase I clinical trial for the treatment of advanced malignancies (clinical trial ID: NCT03440437). By contrast, our current work shows that an Fc domain modified to recognize a BBB receptor can enable CNS delivery of large molecules.

Fig. 1 Modularity and engineering of the TV platform.

(A) The transport vehicle (TV) is a TfR-binding Fc polypeptide that is broadly applicable for protein delivery to the brain. The TV can be fused to Fab arms for bivalent (ATV:BACE1) or bispecific (ATV:BACE1/Tau) targeting, as well as to other proteins, such as lysosomal enzymes (ETV), to facilitate large molecule transport across the BBB. (B) Model of the wild-type Fc homodimer surface (gray; PDB ID: 4W4O) showing the location of the TV library (orange), and regions where FcRn (green outline) and FcɣR1 bind (blue outline). (C) Diagram of the two registers that comprise the TV library in the CH3 domain with sequences of the initial four TVs shown, with the conserved W388 and aromatic421 highlighted. (D) ELISA measuring binding of ATV:BACE1 variants to immobilized hTfR. (E) Immunocytochemistry staining with Alexa488 anti-huIgG1 of 1 μM of either ATVc:BACE1 and anti-BACE1 after incubation with HEK293T cells for 30 min at 37°C. Images are maximum intensity projections of confocal Z-stack images. Cell uptake is quantified by the number of huIgG1-positive puncta per cell. Graph represents mean ± SEM; n = 3 independent experiments; **P < 0.01, unpaired t test; a.u., arbitrary units.

The most well-studied BBB target for brain delivery is the transferrin receptor (TfR), which is highly expressed on brain endothelial cells and undergoes constitutive ligand-independent endocytosis (3438). Early studies with antibodies that bound to rat and mouse TfR showed increases in brain exposure relative to control antibodies, demonstrating the potential of targeting the TfR for BBB transcytosis (39, 40). Although these antibodies accumulated in brain, they were later found to be mostly trapped within the brain endothelium. More recently, low-affinity antibodies that bind monovalently to human TfR (hTfR) were shown to improve brain uptake in mice (24, 41) and nonhuman primates (21). In one example, a bispecific antibody targeting TfR and β-secretase 1 (BACE1) improved CNS exposure in mice and cynomolgus monkeys with a correlative reduction in brain and cerebrospinal fluid (CSF) amyloid β (Aβ) compared to anti-BACE1 antibodies lacking the TfR-targeting moiety (24, 41). Similarly, an anti-Aβ antibody fused at its C terminus to an anti-TfR single-chain Fv (scFv) showed improved brain Aβ reduction in a mouse model of Alzheimer’s disease (AD) as compared to the parental antibody (24). On the basis of these extensive data, we selected TfR as our BBB receptor target for proof-of-concept work in establishing the TV platform. We further selected BACE1 as the therapeutic target for our initial ATVs to exploit soluble Aβ reduction as a simple biomarker for monitoring target engagement in brain and CSF. By combining the TV platform with antibody Fabs that inhibit BACE1, we show that ATVs efficiently cross the BBB and generate a direct pharmacodynamic (PD) response by reducing CNS Aβ production in both hTfR-engineered mice and nonhuman primates. In addition, we demonstrate the modularity of the TV platform by fusing it to two different antibody Fab arms to generate a bispecific ATV, and in a companion paper, we describe its application for an enzyme Fc fusion (42).


Design and screening of TV libraries

Using a crystal structure of a human Fc [Protein Data Bank (PDB) ID: 4W4O], we identified several contiguous patches on the human IgG1 CH3 domain likely amenable to engineering a novel binding epitope. We focused on a register of nine amino acids encompassing about 1000 Å2 of solvent-exposed surface area, distant from both FcRn and FcɣR binding sites (Fig. 1B). This register is distinct from the region previously described for Fcabs and does not contain any insertions or deletions (27). After randomizing the amino acid composition of this register, we transformed the resulting library into yeast for display on the cell surface. Using a combination of magnetic bead sorting and fluorescence-activated cell sorting (FACS), we isolated clones that bound the extracellular domain of the hTfR (fig. S1A). After the final sort round, we identified four unique sequences that shared a conserved tryptophan at position 388 and an aromatic amino acid at residue 421 (Fig. 1C).

We next expressed recombinant versions of the four hTfR-binding Fc domains fused to anti-BACE1 antibody Fab arms (ATVx:BACE1, where “x” designates the TV reference sequence). ATV:BACE1 variants bound to immobilized hTfR by enzyme-linked immunosorbent assay (ELISA; Fig. 1D) and did not compete with native transferrin binding (fig. S1B). Furthermore, we showed that ATV:BACE1 clones bound human embryonic kidney (HEK) 293T cells, which express hTfR but not BACE1, in a dose-dependent manner and internalized after a 30-min incubation at 37°C (Fig. 1E and fig. S1C). Conversely, none of the ATV:BACE1 molecules appreciably bound to cynomolgus monkey TfR (cTfR) (fig. S1D).

Affinity maturation to hTfR and introduction of cTfR cross-reactivity

We used a biased library approach to engineer binding to cTfR and improve affinity to hTfR of the TV sequences (fig. S2, A and B), as cTfR cross-reactivity is critical for subsequent validation of the TV platform in nonhuman primates. Two distinct sequence families retaining the conserved Trp at position 388 and aromatic group at position 421 emerged from the maturation library after several rounds of sorting (table S1). We expressed a representative clone from each family (TV18 and TV35; Fig. 2A) as ATV:BACE1 and assessed their TfR binding properties. Both ATV:BACE1 clones exhibited improved binding to immobilized hTfR and an hTfR-expressing Chinese hamster ovary (CHO) cell line, but did not bind to parental CHO cells (Fig. 2, B and C, and fig. S2C). The matured clones, unlike the parental molecule, also bound cTfR expressed on CHO cells (Fig. 2C).

Fig. 2 Affinity maturation of TV.

(A) Outline of the steps used for TV discovery and affinity maturation. Representative sequences from each step are shown, with original TV library positions highlighted in green. (B and C) Binding of ATV35:BACE1 and ATV18:BACE1 from the soft mutagenesis library, compared to ATVc:BACE1, shows improved affinity to (B) immobilized hTfR and (C) hTfR- and cTfR-overexpressing CHO cells. (D) Affinities of ATV:BACE1 clones from (C) for human and cyno TfRapical, as measured by steady-state surface plasmon resonance, illustrating the increase in affinity to hTfR and cTfR over successive rounds of affinity maturation with clones from the Soft Mutagenesis Library (red), NNK Peripheral Walk Library (black), and Focused Library (blue). MFI, mean fluorescence intensity.

Last, we generated and screened a library designed to interrogate solvent-exposed residues in the CH3 domain outside of the TV epitope to identify peripheral positions that could further improve binding. Using TV35 as a template, we independently randomized 42 positions in CH3 (fig. S3A). We isolated higher-affinity clones from this “NNK Peripheral Walk” library by FACS sorting against a permutated version of the TfR apical domain (TfRapical; fig. S4). The apical domain of TfR was used to increase selection stringency by reducing multivalent binding on the yeast cell surface as compared to dimeric full-length receptor (fig. S3, A and B). Sequences from the selected clones enriched for mutations at six positions, whereas most of the sequence space remained invariant. We explored cooperativity between these six positions using a focused library in which all six residues were concurrently randomized (fig. S5). This library was sorted by FACS against both hTfRapical and cTfRapical, yielding clones with improved binding to both orthologs when reformatted as ATV:BACE1 (Fig. 2, A to D).

In summary, an affinity-matured TV displayed an affinity to hTfRapical of around 100 nM, with corresponding affinity to cTfRapical of about 2 μM. Libraries, representative sequences, and corresponding affinities of ATV:BACE1 clones are summarized in Fig. 2 (A and D).

Crystal structure of ATV35 and hTfRapical cocomplex

To understand the molecular basis of the TV:hTfR interaction, we solved the crystal structure of TV35 bound to hTfRapical to 3.4 Å resolution (Fig. 3A and table S2; PDB ID: 6W3H). The backbone of the TV CH3 domain is relatively unchanged compared to wild-type IgG [PDB ID: 4W4O; Cα root mean square deviation (RMSD): 0.4 Å], indicating that the synthetic epitope does not perturb the IgG fold and that large conformational changes are not required for TfR binding (fig. S6A). In addition, the permutated apical domain overlays well on full-length TfR (PDB ID: 3S91), with a Cα RMSD of 0.9 Å (fig. S4).

Fig. 3 Structural characterization of TV35 in complex with hTfRapical.

(A) Ribbon model of the TV35 Fc dimer (blue and gray) binding two hTfRapical domains (green and wheat). (B) Enlarged view of the binding interface; hTfRapical R208 interacts with TV35 E387. (C) Open book view of the TV35 CH3:hTfRapical binding interface. Residues on hTfRapical within 4 Å of TV are highlighted (orange), and residues on TV35 within 4 Å of TfR are shown as sticks. Positions on TV35 in the mutated register are colored teal, and wild-type positions are colored yellow. S390 is mutated in TV35 but does not participate in binding. (D) Surface model of the TV35:hTfRapical complex overlaid with hIgG1 (PDB ID: 1HZH) and the transferrin:TfR complex (PDB ID: 1SUV). Orientation of TV and Fab arms (blue and dark gray) relative to transferrin (pink and purple) and the cell surface is shown.

Good resolution at the TV-TfR binding interface allows a detailed analysis of the interaction (fig. S6B). The total buried surface area between TV35 and hTfRapical is 1444 Å2 and is derived from eight of the nine mutated residues making direct contacts with hTfRapical (Fig. 3C). Furthermore, Trp388 forms the core of the interaction and is buried in the center of the epitope, offering an explanation for its strict conservation during affinity maturation (Fig. 3C). When overlaid onto the structures of a full human IgG (PDB ID: 1HZH) and a transferrin:TfR complex (PDB ID: 1SUV), TV35 binds the hTfR apical domain in an orientation that appears to project the ATV Fab arms away from both the cell surface and transferrin binding site (Fig. 3D).

The crystal structure of the complex also provides a rationale for the disparate affinities for hTfR and cTfR. Nine amino acids in the apical domain differ between the hTfR and cTfR orthologs, including position 208, which is an arginine in human and glycine in cynomolgus (fig. S7, A and B). Glu387 of TV35 makes a direct contact with Arg208 on hTfR (Fig. 3B), an interaction that is not possible with Gly208 in cTfR. To confirm the importance of this interaction, we displayed variants of hTfR and cTfR apical domains on phage with substituted orthologous mutations. Binding of TV35 was diminished on phage displaying hTfRR208G, whereas binding to cTfRG208R was similar to that observed for hTfR (fig. S7, C and D).

Cellular internalization of ATV:BACE1 is TfR-dependent

For subsequent characterization, we generated ATV:BACE1 clones with a single TfR binding site using knob-into-hole mutations to drive heterodimerization between one IgG heavy chain containing the TV and another lacking the TfR binding site (Fig. 4A) (43). The expressed ATV:BACE1 clones are monodispersed, properly formed heterodimers with favorable thermal stability profiles (fig. S8). ATV variants used for in vivo studies also incorporated L234A/L235A mutations to abrogate FcɣR binding (44), as FcɣR-mediated depletion of circulating reticulocytes is a reported safety liability of TfR binding molecules (45, 46).

Fig. 4 ATV:BACE1 is internalized into human cells in a dose-dependent manner.

(A) Monovalent ATV:BACE1 has a single TfR binding epitope and binds to BACE1 on both Fabs. (B to D) Representative immunocytochemistry images acquired by confocal microscopy shows huIgG1 staining after a 30-min incubation of ATV35.21:BACE1 or anti-BACE1 at 37°C. (B) ATV35.21:BACE1 staining in HEK293T or CHO cells, and anti-BACE1 staining in HEK293T cells. Graphs display mean ± SEM; n = 3 independent experiments; *P < 0.05 for HEK binding of ATV variants compared to anti-BACE1, mixed-model analysis. Scale bars, 50 μm. (C) Staining of ATV35.21:BACE1 in HEK293T cells under permeabilizing and nonpermeabilizing conditions. Graphs display mean ± SEM; n = 2 to 3 independent experiments; ***P < 0.001, unpaired t test. Scale bars, 20 μm. (D) Costaining of ATV:35.21:BACE1 and TfR. Cell uptake is quantified by the number of huIgG1-positive puncta per cell. Graphs display mean ± SEM; n = 3 independent experiments. Scale bars, 50 μm.

ATV:BACE1 variants, but not an anti-BACE1 antibody, were bound and internalized by HEK293T cells in a dose-dependent manner after a 30-min incubation at 37°C, whereas minimal staining was observed in CHO cells that do not express hTfR (Fig. 4B). ATV:BACE1 staining was observed only when HEK293T cells were permeabilized after a 30-min incubation at 37°C, consistent with binding and subsequent internalization of ATV:BACE1 (Fig. 4C). Roughly 50% of intracellular ATV:BACE1 colocalized with hTfR (Fig. 4D), and prolonged exposure of cells to 1 μM ATV:BACE1 or anti-BACE1 did not alter total hTfR expression (fig. S9, A and B). Furthermore, colabeling cellular hTfR and the early endosome marker EEA1 revealed similar colocalization across treatment conditions (fig. S9C), indicating that hTfR trafficking is not affected by ATV:BACE1 binding.

Application of TV to bispecific therapeutic targeting

We fused two different Fab fragments to a TV to modulate distinct pathways involved in the pathology of AD, Aβ production and Tau seeding (ATV:BACE1/Tau; Fig. 5A), to demonstrate the engineering modularity of the TV platform. The resulting ATV:BACE1/Tau molecule was shown using biolayer interferometry to simultaneously bind all three targets: hTfR, BACE1, and Tau (Fig. 5B).

Fig. 5 Example of TV platform applied to a bispecific antibody.

(A) TV can be fused to Fab arms to enable bispecific (BACE1/Tau) targeting. (B) Simultaneous binding of ATV35.21:BACE1/Tau to TfR, BACE1, and Tau as measured by biolayer interferometry. BACE1-loaded biosensors were used to capture ATV35.21:BACE1/Tau and subsequently tested for binding to Tau and hTfR. (C) Percent Aβ40 reduction in media of cultured human neurons after a 24-hour incubation of 33 nM of control IgG, anti-Tau, anti-BACE1, or ATV35.21:BACE1/Tau, in the presence or absence of 50 nM of recombinant Tau-GFP (green fluorescent protein). Data were normalized to percent of untreated control neurons. Graphs represent mean ± SEM; n = 3 biological replicates. ****P < 0.0001, two-way ANOVA compared to untreated. (D) FRET signal from Tau biosensor FRET cells after incubation with 267 nM of control IgG, anti-BACE1, anti-Tau, or ATV35.21:BACE1/Tau for 24 hours, in the presence of healthy (control) or Alzheimer’s disease (AD) human brain lysate. Graph represents mean ± SEM; n = 3 biological replicates; ***P < 0.001, two-way ANOVA compared to untreated.

We next addressed whether each Fab of ATV:BACE1/Tau maintains the ability to engage its respective target in biologically relevant cellular assays. Incubation of ATV:BACE1/Tau with primary human cortical neuron cultures expressing endogenous hTfR and BACE1 resulted in a ~50% reduction in Aβ concentrations relative to untreated neurons, whereas no reduction in Aβ amounts was observed in either control IgG–treated or anti-Tau–treated cells (Fig. 5C). The addition of excess recombinant monomeric human Tau did not affect anti-BACE1 activity of the ATV bispecific (Fig. 5C).

We used a fluorescence resonance energy transfer (FRET)–based cellular model of Tau seeding using HEK293T cells to evaluate the effectiveness of ATV:BACE1/Tau in reducing Tau aggregation from AD brain lysate (47). Across treatment conditions, AD patient brain lysate induced an increase in FRET signal compared to control brain lysate, indicative of intracellular Tau seeding and aggregation (Fig. 5D). ATV:BACE1/Tau inhibited AD lysate-induced Tau aggregation to an equivalent degree as a bivalent anti-Tau antibody compared to untreated cells, whereas both control IgG and anti-BACE1 antibodies had no effect (Fig. 5D). Together, these results demonstrate that the modular nature of the TV platform enables functional multispecific therapeutic target binding within a native IgG scaffold.

TV increases brain uptake of anti-BACE1 Fabs in TfRmu/hu KI mice

We next set out to establish in vivo proof of concept for our TV platform by generating a knock-in (KI) mouse that could be used to study target engagement, since the TVs specifically bind to primate but not murine TfR (fig. S2C). Our approach was to create a chimeric TfRmu/hu KI mouse harboring the hTfR apical domain knocked into the murine TfR, rather than replacing the full-length receptor. The resulting TfRmu/hu KI mouse preserves the native murine transferrin ligand binding domain and retains the majority of Tfrc introns, minimizing potential dysregulation of protein expression and function. Consistent with this approach, total TfR protein expression in brain and peripheral tissues of homozygous TfRmu/hu KI mice and wild-type mice was indistinguishable (fig. S10, A to D).

We first dosed the TfRmu/hu KI mice with either an anti-BACE1 antibody or ATV:BACE1 to determine whether the TV platform enables brain uptake of an intravenously administered molecule. Twenty-four hours after a single dose (50 mg/kg), ATV:BACE1 brain exposure was nearly 40-fold higher than anti-BACE1 (38 and 1 nM, respectively) (Fig. 6A). ATV:BACE1 demonstrated substantial target engagement in brain, reducing soluble brain Aβ40 by 57% as compared to anti-BACE1–treated animals (Fig. 6B).

Fig. 6 PKPD and brain distribution of ATV:BACE1 in TfRmu/hu KI mice.

(A and B) Brain huIgG1 (A) and brain mouse Aβ40 concentrations (B) 24 hours after a single intravenous dose (50 mg/kg) of anti-BACE1 or ATV:BACE1. Graphs represent mean ± SEM; n = 8 per group; ****P < 0.0001 using unpaired t test compared to anti-BACE1. (C to F) Plasma (C), brain (D) concentrations, and brain-to-plasma ratio (E) of anti-BACE1 and ATV:BACE1 variants, and brain mouse Aβ40 concentrations (F) at 1, 3, and 7 days after a single intravenous injection (50 mg/kg) in TfRmu/hu KI mice. Graphs represent mean ± SEM; n = 5 per group; **P < 0.01 and ****P < 0.0001, based on linear mixed-model analysis. Dashed line represents lower limit of quantification (LLOQ) (C). (G and H) Confocal imaging of sections from TfRmu/hu KI mice brain cortex 4 and 24 hours after a dose of either anti-BACE1 or ATV35.21:BACE1 (50 mg/kg). Sections were immunostained with antibodies against huIgG1 (purple) or the neuronal marker NeuN (cyan). Arrows indicate huIgG1 staining localized to putative vascular profiles (G). Arrowheads indicate neuronal huIgG1 staining (H). Dashed boxes in merge panels indicate regions shown at higher magnification with super-resolution confocal imaging displayed at far right, counterstained with 4′,6-diamidino-2-phenylindole (DAPI) dye to indicate non-neuronal cell nuclei (white; for example, vasculature-associated endothelial cells) in addition to huIgG1 and NeuN immunoreactivity.

We next evaluated brain and plasma exposure after a single dose (50 mg/kg, intravenously) of two representative ATV:BACE1 clones that differed in their binding affinity to hTfR (Fig. 2D). Both variants exhibited higher plasma clearance in an affinity-dependent manner compared to anti-BACE1, consistent with TfR-mediated disposition in the periphery (Fig. 6C). Both molecules had enhanced brain exposure at 1 and 3 days after dose compared to anti-BACE1 (Fig. 6D). Depending on the time point, ATV:BACE1 variants exhibited about 10- to 100-fold greater brain-to-plasma ratios than anti-BACE1, illustrating how TfR binding improves brain transport efficiency (Fig. 6E). The brain exposure of ATV:BACE1 resulted in a sustained reduction of brain Aβ40 concentration compared to anti-BACE1–treated mice after a single dose (Fig. 6F). Imaging of brain sections immunostained for human IgG1 (huIgG1) demonstrated a qualitative transition in staining from a predominantly vascular pattern at 4 hours after ATV:BACE1 dosing to a more broadly diffuse parenchymal pattern with prominent cellular internalization at 24 hours after ATV:BACE1 dosing in cortex (fig. S11, A, D, and E), hippocampus (fig. S11, A, H, and I), and cerebellum (fig. S11, A, L, and M). In contrast, brain tissue from mice administered anti-BACE1 exhibited markedly less huIgG1 immunoreactivity at both 4 and 24 hours after dose in cortex (fig. S11, A to C), hippocampus (fig. S11, A, F, and G), and cerebellum (fig. S11, A, J, and K), consistent with the low concentration of this antibody measured in brain (Fig. 6, A and D). Subcellular localization of huIgG1 in the brains of mice dosed with ATV:BACE1 or anti-BACE1 was further examined using super-resolution confocal imaging in brain sections from cortex (Fig. 6, G and H) and hippocampus (fig. S11, N and O). Imaging at subcellular resolution confirmed (i) an ATV:BACE1 staining pattern that exhibited intense, punctate vascular profiles suggestive of robust brain endothelial cell uptake at 4 hours after dose; (ii) a more diffuse pattern of ATV:BACE1 staining in the neuropil with prominent internalization into NeuN-positive neurons at 24 hours after dose; (iii) no obvious vascular signal for anti-BACE1 suggestive of minimal brain endothelial cell uptake at 4 hours after dose; and (iv) much lower anti-BACE1 signal in the neuropil and within NeuN-positive neurons at 24 hours after dose. Quantitative analysis of Fig. 6 (G and H) of total huIgG1 immunoreactivity from super-resolution confocal microscopy (SRCM) confirmed low anti-BACE1 signal in cortex that was just a small fraction of the total signal present in cortical fields from animals dosed with ATV:BACE1 (fig. S12A), consistent with lower-resolution imaging of larger sections of the cortex (fig. S11, B to E). SRCM imaging results suggested about 30-fold higher total ATV:BACE1 signal than anti-BACE1 signal in cortex at 24 hours, in rough agreement with results based on brain lysate (Fig. 6A). Because much of the cortical anti-BACE1 signal may be attributable to background autofluorescence, we performed a background subtraction on ATV:BACE1 quantitative data using anti-BACE1 huIgG1 signal values colocalized with NeuN (neuronal background) and remaining anti-BACE1 huIgG1 signal (non-neuronal background) to estimate specific neuronal versus non-neuronal signal at 4 and 24 hours for ATV:BACE1 (fig. S12B). The results confirmed predominantly non-neuronal signal (including vascular profiles) associated with ATV:BACE1 at 4 hours (~75% of total) that transitioned to approximately equal neuronal and non-neuronal signal fractions at 24 hours.

High-affinity, but not low-affinity, TfR molecules have been previously found to reduce brain TfR expression after treatment in mice (48). Total brain TfR protein expression was indistinguishable between different groups of TfRmu/hu KI mice after a single-dose (50 mg/kg) treatment of anti-BACE1 or ATV:BACE1 variants (fig. S13A), in line with lower-affinity ATV binding to TfR. Furthermore, brain TfR expression remained unaltered in chronic dosing conditions compared to after a single dose, indicating that TfR expression is not affected after multidosing of ATV (fig. S13B). To more specifically evaluate potential TfR expression changes at the BBB, as opposed to in whole brain, we separated mouse brain vascular and parenchymal fractions using the capillary depletion method in TfRmu/hu KI mice. Immunoreactivity of CD31 (an established brain endothelial cell marker) was observed exclusively in the vascular fraction, but not in the parenchymal lysate fraction (P < 0.05; fig. S13C), confirming successful separation of the two brain compartments. TfR protein expression was 1.85-fold higher in the vascular fraction than in the parenchymal lysate fraction (P < 0.05), although immunoreactivity was also observed in the brain parenchymal fraction, consistent with reports of wide TfR expression on neurons in the adult mouse CNS (fig. S13C) (49). Furthermore, TfR expression in the vascular fraction was comparable between untreated and ATV:BACE1-treated mice, indicating that the TfR population with the highest exposure to circulating ATVs was not affected (fig. S13D).

We next evaluated plasma PK of ATV:BACE1 in wild-type mice to determine whether the higher plasma clearance observed for ATV:BACE1 variants in TfRmu/hu KI mice was mediated by TfR binding in the periphery. In the absence of hTfR binding, ATV:BACE1 variants exhibited clearance comparable to anti-BACE1 after a single dose (10 mg/kg, intravenously), and no TfR-mediated disposition was observed (fig. S14A). These observations are consistent with a lack of ATV binding to the rodent TfR; they also indicate that the TV mutations preserve FcRn recycling and maintain in vivo stability without off-target binding over the time course of the experiment.

TV enhances anti-BACE1 brain exposure and PD response in nonhuman primates

To validate the TV platform in nonhuman primates, we evaluated ATVs in cynomolgus monkeys, which also express TfR at the brain vasculature (fig. S14B). Monkeys were dosed with control IgG, anti-BACE1, or ATV35.21.16:BACE1 (30 mg/kg, intravenously), which binds to cTfR with a KD (dissociation constant) of 1.9 μM (Fig. 2D). We monitored serum huIgG1, plasma and CSF Aβ40, soluble amyloid precursor protein β (APPβ), and APPα concentrations over 4 weeks. Compared to anti-BACE1 and control IgG, ATV35.21.16:BACE1 exhibited higher systemic clearance, consistent with TfR-mediated disposition (Fig. 7A). Treatment with either anti-BACE1 or ATV35.21.16:BACE1 elicited a reduction in plasma Aβ40 concentrations compared to control IgG (Fig. 7B). Although both groups achieved similar degrees of maximal inhibition, anti-BACE1–treated animals achieved a longer duration of Aβ40 reduction in plasma, presumably due to higher overall exposure over time, as compared to ATV35.21.16:BACE1 (Fig. 7, A and B).

Fig. 7 PKPD of ATV:BACE1 in nonhuman primates.

(A to D) Serum huIgG1 concentration (A), plasma Aβ40 (B), CSF Aβ40 (C), and CSF soluble APPβ/α ratio (D) in cynomolgus monkeys over 4 weeks after a single intravenous dose of control IgG, anti-BACE1, or ATV35.21.16:BACE1 (30 mg/kg). Graphs display mean ± SEM; n = 4 to 5 per group; ***P < 0.001 and ****P < 0.0001 based on a linear mixed-model analysis. (E to G) huIgG1 (E) and soluble APPβ/α ratio (F) from various cynomolgus monkey brain regions and immunohistochemical staining from cynomolgus monkey cortex (G) 48 hours after a single intravenous dose of control IgG, anti-BACE1, or ATV35.21.16:BACE1 (30 mg/kg). Graphs display mean ± SEM; n = 3 to 4 per group; *P < 0.05, **P < 0.01, and ****P < 0.0001 using two-way ANOVA compared to anti-BACE1. n.s., not significant.

Longitudinal serial sampling of CSF allowed for monitoring of Aβ40 concentration over the duration of the study. Cynomolgus monkeys treated with ATV35.21.16:BACE1 exhibited up to a 70% reduction in CSF Aβ40 and a 75% reduction in soluble APPβ/APPα concentration ratio [which accounts for biological variability in APP expression (50)] 2 days after dosing (Fig. 7, C and D). The CSF Aβ concentration returned to baseline by day 14 as peripheral exposure diminished (Fig. 7A). Conversely, there was no reduction in CSF Aβ or the soluble APPβ/APPα concentration ratio in animals dosed with control IgG or anti-BACE1 (Fig. 7, C and D). We tested all groups for amount of circulating blood reticulocytes and serum iron and determined that these parameters were unchanged over the course of the study (fig. S14, C and D) (45, 46).

In a subsequent study, we dosed cynomolgus monkeys with control IgG, anti-BACE1, or ATV35.21.16:BACE1 (30 mg/kg, intravenously). Animals were sacrificed after 48 hours, and brain concentrations of huIgG1 and sAPPβ/α ratio were evaluated. Across multiple brain regions, the concentrations of ATV:BACE1 were 26- to 35-fold higher than that of anti-BACE1 (Fig. 7E). With these increased brain concentrations of ATV:BACE1, we observed a 32 to 69% reduction in sAPPβ/α as compared to control IgG, depending on the brain region (Fig. 7F). Monkeys treated with anti-BACE1 had little accumulation of huIgG1 in brain and had sAPPβ/α concentration ratios that were indistinguishable from animals administered control IgG (Fig. 7F). Immunohistochemistry of the ventral cortex (Fig. 7G) was qualitatively consistent with the ELISA-based quantification of huIgG1 concentrations (Fig. 7E): Sections from animals dosed with ATV:BACE1 exhibited broad parenchymal huIgG staining in contrast to minimal staining in sections from animals dosed with anti-BACE1.

Last, we compared the efficiency of brain uptake across species (Fig. 8) using PK data from both the mouse (Fig. 6, A and C) and monkey (Fig. 7E) for ATV:BACE1 and control antibodies at brain Cmax (24 and 48 hours after dose for mouse and monkey, respectively). The percent brain-to-plasma huIgG1 concentration ratio was ninefold higher in mice dosed with ATV:BACE1 than for mice dosed with anti-BACE1 control antibody 24 hours after dose (P < 0.01). In contrast, the percent brain-to-plasma huIgG1 concentration ratio for ATV:BACE1 was over 27-fold higher than for control IgG in monkeys 48 hours after dose (P < 0.0001). No notable differences were observed in the percent CSF-to-plasma huIgG1 concentration ratios between monkeys dosed with ATV:BACE1 and those dosed with control antibodies 48 hours after dose, consistent with increased ATV:BACE1 brain exposure having resulted from enhanced penetration across the BBB.

Fig. 8 Comparison of plasma and brain exposures across species.

Percent CSF and brain-to-plasma ratios for intravenously dosed control IgG, anti-BACE1, and ATV:BACE1 in mouse (data from Fig. 6, A and C) and monkey (data from Fig. 7E). ****P < 0.0001 compared to control IgG based on a linear mixed-model analysis for cynomolgus monkey data (brain/plasma data points represent different brain regions per animal from n = 4 study as described in Fig. 7); **P < 0.01 compared to anti-BACE1 based on paired t test for mouse data (data points for ATV:BACE1 include ATV35.21:BACE1 and ATV35.BACE1 from study described in Fig. 6).


The past 30 years of drug discovery have yielded efficacious therapeutic antibodies and proteins for the treatment of many debilitating pathologies including cancer, inflammatory diseases, and rare genetic disorders (51). Unfortunately, the development of disease-modifying therapies for neurological indications, including AD and Parkinson’s diseases, has largely been absent from such progress. Many factors contribute to the ultimate success or failure of therapeutic candidates, and drug development to treat neurological conditions is strongly affected by the inability of many biotherapeutics to achieve efficacious concentrations in the CNS after peripheral administration (7).

The present study introduces the development of a highly modular TV platform that binds a brain endothelial cell–enriched target (TfR) to enable brain delivery of biotherapeutics. We demonstrate enhanced brain uptake and PD activity in both mouse and nonhuman primates using quantitative PKPD analysis and high-resolution qualitative immunostaining. The TV platform may be engineered to accommodate other advantageous configurations, including bispecific ATVs exhibiting native IgG size and geometry (for example, ATV:BACE1/Tau, as demonstrated here) and protein fusions smaller than conventional protein-antibody fusions. We also introduce the ability to apply the platform to enzyme-Fc fusions with the enzyme TV (ETV), described in a companion paper (42).

For initial platform validation, we generated TVs against the TfR because conventional antibody approaches targeting this receptor have repeatedly demonstrated enhanced brain uptake in both rodent and nonhuman primate models (21, 24, 41, 5254). Our initial challenge was to introduce TfR binding into the Fc domain of huIgG1. The Fc largely serves as a stable structural scaffold that binds to Fcγ receptors near the hinge to elicit effector functions, as well as FcRn, complement receptor C1q, and TRIM21 near the CH2-CH3 interface to enable endosomal recycling, complement activation, and cytoplasmic degradation, respectively (55). Conversely, the CH3 domain of the Fc does not have any known intrinsic interactions (27), making it an attractive target for engineering novel receptor binding without interfering with the natural Fc functions.

Engineering new molecular recognition into previously existing binding interfaces has been accomplished now for several well-established scaffold proteins (56). However, engineering an inert protein surface to interact with a protein of interest remains a substantial challenge. To accomplish this, we built a number of libraries in which contiguous surfaces formed by solvent-exposed side chains were randomized and screened for binding to TfR. One library, located well away from the FcRn binding site, produced several sequences that bound weakly to hTfR yet also displayed little cross-reactivity to cTfR. Over several rounds of affinity maturation, a TV sequence was engineered with affinities to hTfR and cTfR of 120 and 1900 nM, respectively.

TVs were initially fused to anti-BACE1 Fabs to enable the use of soluble Aβ40 as a validated CNS PD readout in mice and cynomolgus monkeys. We also demonstrate the advantage of having both Fab arms available for therapeutic targeting by generating a bispecific TV that targets both BACE1 and Tau, two sources of pathology in AD patients (57). ATV:BACE1/Tau can simultaneously engage all three targets and inhibits both Aβ40 formation and Tau spreading in vitro. The ability to bind two different Fab targets provides unique opportunities for designing combination therapies within a native IgG structure. In addition, because protein size may affect further biodistribution within the CNS after crossing the BBB (58), TVs may have advantages for multispecific therapeutics compared to larger, more traditional monoclonal antibody (mAb) fusion proteins larger than native IgG.

When TfR-targeted ATV:BACE1 variants were compared to a standard anti-BACE1 antibody in vivo, ATVs yielded about 10- to 40-fold higher brain IgG concentrations in rodents 24 to 72 hours after administration and >30-fold higher IgG concentrations in multiple regions across the brain of cynomolgus monkeys 48 hours after administration. Furthermore, ATV:BACE1 reduced brain and CSF Aβ concentrations in both species across a wide range of TfR affinities (from 120 nM in mouse to 1900 nM in monkey), whereas the standard anti-BACE1 yielded no appreciable CNS PD effects. Brain sections from both rodents and cynomolgus monkeys showed broad huIgG1 immunoreactivity within the parenchyma of ATV-dosed animals that contrasted markedly with the minimal huIgG1 staining observed for anti-BACE1–dosed animals. Epifluorescence microscopy and SRCM at subcellular resolution demonstrated a predominantly cerebrovasculature-associated ATV:BACE1 biodistribution across multiple brain areas in the mouse at an early time point (4 hours after dose), similar to that described in earlier studies when conventional mAbs binding TfR with high affinity were administered peripherally (59). However, imaging at a later time point corresponding to brain Cmax in the mouse (24 hours after dose) revealed a transition to a robust parenchymal signal for ATV:BACE1, including internalization into NeuN-positive neurons, in contrast to the typically faint parenchymal signal seen 24 hours after dose with high-affinity TfR binding mAbs and antibody fusion proteins (25, 59). Transition from an early microvascular endothelial cell internalization pattern at 4 hours to a parenchymal pattern with prominent neuronal internalization at 24 hours is most consistent with ATV reaching the brain following passage across brain endothelial cells, as opposed to other mechanisms such as ATV crossing the blood-CSF barriers and subsequent transport into the brain from the CSF [as has been described for control IgG; (60)]. Widespread ATV biodistribution was markedly different than the huIgG1 immunoreactivity pattern observed for anti-BACE1–treated control animals, where signal was limited mostly to the pial brain surface. The mechanism by which ATVs are transported across the BBB is likely dependent on endosomal trafficking routes. One possibility is that the ATV/TfR co-complex is transported to the basolateral side of brain endothelial cells, where its weak interaction would promote release of the ATV into the brain parenchyma. Alternately, ATVs may dissociate from TfR within brain endothelial cell endosomes and be further trafficked via FcRn or some other receptor to either the apical or basolateral membrane. The precise steps involved will require further study.

It is well understood that extrapolations between brain and CSF antibody concentrations have numerous pitfalls (61). Oft-quoted ~0.1% antibody exposures in the CNS of human subjects and rodents typically refer to CSF-to-plasma ratios, not brain-to-plasma ratios (62). Specific measurements of brain-to-plasma exposure for nontargeted antibodies have yielded ratios far lower than 0.1% (11), suggesting that CSF typically overrepresents brain exposure for such proteins. As brain-to-plasma and CSF-to-plasma ratios indicate drug exposure independent of dose, we used these data to infer potential mechanisms of antibody biodistribution to the CNS. Simplistically, if transport from plasma to brain across the BBB is dominant, then brain exposure might exceed CSF exposure. Conversely, if transport from plasma to CSF across the blood-CSF barriers is dominant (or if BBB transport is minimal), then CSF exposure will be expected to exceed brain exposure. Several key points emerge from a comparison of CNS exposure ratios for our studies. First, the average brain:plasma ratio for ATV:BACE1 was over 20-fold greater than control antibodies across studies. Second, the brain:plasma ratio was over fourfold greater than the CSF-to-plasma ratio for ATV:BACE1 in cynomolgus monkeys, consistent with dominant brain penetration across the BBB. Third, both control antibodies exhibited substantially lower brain:plasma ratios than their corresponding CSF-to-plasma ratios in monkeys, consistent with minimal brain penetration across the BBB; CSF concentrations of control antibodies therefore overestimated true brain exposure, in line with previous reports for circulating nontargeted IgG (11, 63). Anti-BACE1 brain:plasma ratios were ninefold lower in monkeys than in mice, suggesting that circulating nontargeted IgG may be even less readily accessible to the brain in nonhuman primates. Together, the results emphasize the advantage of the TV platform to achieve considerable brain exposure for antibody therapeutics across species, with brain ATV:BACE1 concentrations exceeding CSF concentrations at Cmax.

Having demonstrated the utility of TfR-targeted TVs for increasing the brain uptake of biotherapeutics, future efforts will focus on applying this platform to the development of therapeutic ATVs, PTVs, and ETVs for the treatment of neurodegenerative diseases. Several factors will need to be the subject of future studies to address limitations of our current work. While ATV:BACE1 molecules showed broad uptake and distribution throughout the brain in cynomolgus monkeys, the disparate affinities between hTfR and cTfR confounds the interpretation of how ATVs will perform in humans. In addition, safety of TfR targeting for chronic dosing, immunogenicity of the molecule due to introduced mutations, and how the combination of TfR and therapeutic Fab targeting may affect biodistribution and downstream biological effects in both the nervous system and the periphery need to be explored. Ultimate proof of concept will be in the clinic, where hTfR-targeted TV platform variants can be evaluated against precisely the epitopes they have been engineered toward. Last, it will be of great interest to engineer TVs that bind other BBB targets, with the goal of identifying additional therapeutics with transcytosis and biodistribution properties different from the TfR-targeted ATVs described in the present study.


Study design

This study was designed to engineer and characterize the TV, an Fc domain engineered to bind to the TfR to enable brain delivery of biotherapeutics. First, we engineered the TV by initially generating various libraries, wherein contiguous positions on the Fc were randomized, and then sorting those libraries to isolate TfR binders. The biophysical and biochemical properties of the TfR binding TV clones were characterized, and the sequences underwent several rounds of affinity maturation to improve affinities to both hTfR and cTfR. Last, we studied the biodistribution, brain penetrance, and safety of the resulting TV sequences as ATVs in both a humanized TfR mouse and cynomolgus monkey.

Sample sizes for in vivo studies were determined on the basis of statistical power calculations from previous studies as well as our experiences with the models and endpoints used. Figure legends contain sample sizes, replicate information, and statistical tests used. For all in vivo studies, animals were randomly assigned to groups before experiments, and researchers were blinded to treatment when possible during quantification and analysis.

Statistical analysis

Data have been expressed as mean ± SEM as indicated in figure legends. All statistical analysis was performed in GraphPad Prism 8. Analysis was done using two-way analysis of variance (ANOVA), unpaired t test, paired t test, or linear mixed-model analysis, as indicated in figure legends. Criterion for differences to be considered significant was P < 0.05.


Materials and Methods

Fig. S1. Initial TV library sorting and characterization of first-generation ATV:BACE1 clones.

Fig. S2. Affinity maturation of initial TV clones by soft mutagenesis.

Fig. S3. Affinity maturation of TV using a peripheral NNK walk.

Fig. S4. Design of a circularly permutated TfR apical domain.

Fig. S5. Affinity maturation of TV using a focused, peripheral library.

Fig. S6. Structural analysis of TV35 and TfRapical.

Fig. S7. TfRapical point mutations displayed on phage reveal the residues affecting affinity differences between hTfR and cTfR.

Fig. S8. Biophysical characterization of knob-in-hole ATV:BACE1 variants.

Fig. S9. TfR expression and subcellular localization are not affected by ATV:BACE1 incubation in HEK293T cells.

Fig. S10. Characterization of TfR expression in TfRmu/hu KI mice.

Fig. S11. Immunohistochemical localization of anti-BACE1 and ATV:BACE1 in TfRmu/hu KI mice brain regions after a dose (50 mg/kg).

Fig. S12. Quantification of huIgG1 in TfRmu/hu KI mouse cortex from SRCM.

Fig. S13. Characterization of brain TfR expression after dose.

Fig. S14. PK of ATV:BACE1 variants in WT mice and safety of ATV:BACE1 in cynomolgus monkeys.

Table S1. Representative TV sequences isolated from soft mutagenesis affinity maturation library.

Table S2. Crystallographic parameters for the TV35:TfRapical structure.

Data file S1. Raw data.

References (6466)


Acknowledgments: We thank the protein engineering scientists at F-star Biotechnology, including R. Fiehler and M. Tuna, for helpful discussions on Fc engineering and data analysis. We thank S. Mosesova for statistical analysis support. We thank K. Fertig and colleagues at Leica Microsystems for assistance with super-resolution confocal imaging. Funding: This study was funded by Denali Therapeutics. Author contributions: Conceptualization: M.S.K., R.C.W., R.J.W., M.S.D., R.G.T., A.P.S., and Y.J.Y.Z.; formal analysis: J.A.G. and K.R.H.; investigation: M.S.K., R.C.W., J.A.G., W.K., C.S.M., R.T., D.J.K., A.S., C.B., K.R.H., T.G., V.A.A., X.C., Y.Z., H.S., K.J., L.K., T.M., K.S.C., M.E.P., N.L., M.E.K.C., S.L.D., S.B., A.P.S., and Y.J.Y.Z.; writing (original draft): M.S.K., R.C.W., A.P.S., and Y.J.Y.Z.; writing (review and editing): M.S.K., R.C.W., R.G.T., A.P.S., and Y.J.Y.Z.; supervision: M.S.K., P.E.S., S.H., Z.K.S., R.G.T., R.J.W., A.P.S., and Y.J.Y.Z. Competing interests: M.S.K., R.C.W., J.A.G., W.K., C.S.M., R.T., D.J.K., A.S., C.B., K.R.H., T.G., V.A.A., X.C., Y.Z., H.S., K.J., P.E.S., L.K., T.M., K.S.C., M.E.P., N.L., M.E.K.C., S.L.D., S.B., S.H., Z.K.S., R.G.T., R.J.W., M.S.K., A.P.S., and Y.J.Y.Z. are paid employees of Denali Therapeutics Inc. Denali has filed patent applications related to the subject matter of this paper; WO 2018/152375, ENGINEERED POLYPEPTIDES; WO2018/152285, TRANSFERRIN RECEPTOR TRANSGENIC MODELS; and WO 2018/152326, ENGINEERED TRANSFERRIN RECEPTOR BINDING POLYPEPTIDES. R.G.T. currently serves on scientific advisory boards for Alcyone Lifesciences Inc. and a Lundbeck Foundation supported Research Initiative on Brain Barriers and Drug Delivery. Data and materials availability: All data associated with this study are present in the paper or in the Supplementary Materials.

Stay Connected to Science Translational Medicine

Navigate This Article