Predicting the clinical subcutaneous absorption rate constant of monoclonal antibodies using only the primary sequence: a machine learning approach.

Subcutaneous injections are an increasingly prevalent route of administration for delivering biological therapies including monoclonal antibodies (mAbs). Compared with intravenous delivery, subcutaneous injections reduce administration costs, shorten the administration time, and are strongly preferred from a patient experience point of view. An understanding of the absorption process of a mAb from the injection site to the systemic circulation is critical to the process of subcutaneous mAb formulation development. In this study, we built a model to predict the absorption rate constant (ka), which denotes how fast a mAb is absorbed from the site of administration. Once trained, our model (enabled by the XGBoost algorithm in machine learning) can predict the ka of a mAb following a subcutaneous injection using in silico molecular properties alone (generated from the primary sequence). Our model does not need clinically observed plasma concentration-time data; this is a novel capability not previously achieved in predictive pharmacokinetic models. The model also showed improved performance when benchmarked against a recently reported mechanistic model that relied on clinical data to predict subcutaneous absorption of mAbs. We further interpreted the model to understand which molecular properties affect the absorption rate and showed that our findings are consistent with previous studies evaluating subcutaneous absorption through direct experimentation. Taken altogether, this study reports the development, validation, benchmarking, and interpretation of a model that can predict the clinical ka of a mAb using its primary sequence as the only input.

Clinical Investigation of Large Volume Subcutaneous Delivery up to 25 mL for Lean and Non-Lean Subjects.

PURPOSE: To evaluate the clinical feasibility and tolerability of large volume subcutaneous delivery at different injection depths for lean and non-lean subjects. METHODS: A single-center, randomized, subject-blinded, crossover study in 62 healthy subjects was conducted to evaluate delivery of a 10-cP solution containing hyaluronic acid. Subjects were separated into lean and non-lean cohort by SC thickness. A syringe pump was used to study the effect of different volumes (5, 12, 25 mL) of a viscous placebo solution and needle lengths (6, 9 and 12 mm) delivered at 0.5 mL/min. RESULTS: Across all treatments, injection sites were observed to have negligible leakage, ~34 kPa of back pressure, and VAS of mild pain with higher pain from needle insertion than during injection. While mild to moderate erythema was the most frequently reported ISR and edema was most prominent for 25 mL injections, all ISRs were resolved within 4 hours post injection. Subjects were unbothered by ISRs across all treatments and rated them as low distress scores (average 1.0-1.5 out of 6). CONCLUSION: SC injection of 25 mL is feasible and tolerable using a low-pain formulation for abdomen injection irrespective of subcutaneous thickness and injection depths at a delivery rate of 0.5 mL/min.

Concentration-dependent diffusion of unlabeled protein within an in vitro hyaluronic acid matrix.

Diffusion and movement of subcutaneously injected biologics and high-concentration immunoglobulin G (IgG) therapeutics away from the injection site and through the subcutaneous (SC) tissue may be concentration dependent. This possibility was confirmed by in situ measurement of diffusion coefficients of unlabeled bovine IgG in phosphate-buffered saline within an in vitro hyaluronic acid matrix that represents the SC electrostatic environment. Diffusion decreased from 2.67 to 0.05 × 10-7 cm2 /s when IgG concentration increased from 25 to 73 mg/mL. The results demonstrated that in situ detection of unlabeled proteins within an in vitro SC environment provides another useful tool for the preclinical characterization of injectable biologics.

In vitro measurement of concentration of unlabeled protein within a hyaluronic acid matrix.

There are currently more than 560 therapeutic monoclonal antibodies (mAbs) at various stages of research and clinical testing, including candidates for administration by subcutaneous (SC) injection. Preclinical studies based on in vitro measurements of high molecular weight proteins within simulated SC matrices are assisting laboratory studies of interactions of injectable biotherapeutic proteins within the SC environment in relation to bioavailability. We report a new method for directly measuring diffusion of unlabeled, high molecular weight proteins injected into an in vitro matrix that simulates the negatively charged environment of the SC. The matrix consists of 10 mg/ml HA in a repurposed cell culture chamber. The measurement consists of pipetting triplicate 20 µl protein samples into the matrix, placing the chamber in a laboratory scanner, activating tryptophan residues in the protein at 280 nm, and imaging the resulting protein fluorescence at 384 nm over a 0.5-4 h time period thus tracking protein movement. This facile approach enables mapping of protein concentration as a function of time and distance within the matrix, and determination of diffusion coefficients, D, within ±10%. Bovine IgG and BSA gave D = 2.3 ± 0.2*10-7 and 4.6 ± 0.2*10-7 cm2 /s at 24°C, respectively, for initial protein concentrations of 21 mg/mL.

Engineered ligand-based VEGFR antagonists with increased receptor binding affinity more effectively inhibit angiogenesis.

Pathologic angiogenesis is mediated by the coordinated action of the vascular endothelial growth factor (VEGF)/vascular endothelial growth factor receptor 2 (VEGFR2) signaling axis, along with crosstalk contributed by other receptors, notably αvß3 integrin. We build on earlier work demonstrating that point mutations can be introduced into the homodimeric VEGF ligand to convert it into an antagonist through disruption of binding to one copy of VEGFR2. This inhibitor has limited potency, however, due to loss of avidity effects from bivalent VEGFR2 binding. Here, we used yeast surface display to engineer a variant with VEGFR2 binding affinity approximately 40-fold higher than the parental antagonist, and 14-fold higher than the natural bivalent VEGF ligand. Increased VEGFR2 binding affinity correlated with the ability to more effectively inhibit VEGF-mediated signaling, both in vitro and in vivo, as measured using VEGFR2 phosphorylation and Matrigel implantation assays. High affinity mutations found in this variant were then incorporated into a dual-specific antagonist that we previously designed to simultaneously bind to and inhibit VEGFR2 and αvß3 integrin. The resulting dual-specific protein bound to human and murine endothelial cells with relative affinities of 120 ± 10 pM and 360 ± 50 pM, respectively, which is at least 30-fold tighter than wild-type VEGF (3.8 ± 0.5 nM). Finally, we demonstrated that this engineered high-affinity dual-specific protein could inhibit angiogenesis in a murine corneal neovascularization model. Taken together, these data indicate that protein engineering strategies can be combined to generate unique antiangiogenic candidates for further clinical development.

Inhibition of the GAS6/AXL pathway augments the efficacy of chemotherapies.

The AXL receptor and its activating ligand, growth arrest-specific 6 (GAS6), are important drivers of metastasis and therapeutic resistance in human cancers. Given the critical roles that GAS6 and AXL play in refractory disease, this signaling axis represents an attractive target for therapeutic intervention. However, the strong picomolar binding affinity between GAS6 and AXL and the promiscuity of small molecule inhibitors represent important challenges faced by current anti-AXL therapeutics. Here, we have addressed these obstacles by engineering a second-generation, high-affinity AXL decoy receptor with an apparent affinity of 93 femtomolar to GAS6. Our decoy receptor, MYD1-72, profoundly inhibited disease progression in aggressive preclinical models of human cancers and induced cell killing in leukemia cells. When directly compared with the most advanced anti-AXL small molecules in the clinic, MYD1-72 achieved superior antitumor efficacy while displaying no toxicity. Moreover, we uncovered a relationship between AXL and the cellular response to DNA damage whereby abrogation of AXL signaling leads to accumulation of the DNA-damage markers γH2AX, 53BP1, and RAD51. MYD1-72 exploited this relationship, leading to improvements upon the therapeutic index of current standard-of-care chemotherapies in preclinical models of advanced pancreatic and ovarian cancer.

An engineered Axl 'decoy receptor' effectively silences the Gas6-Axl signaling axis.

Aberrant signaling through the Axl receptor tyrosine kinase has been associated with a myriad of human diseases, most notably metastatic cancer, identifying Axl and its ligand Gas6 as important therapeutic targets. Using rational and combinatorial approaches, we engineered an Axl 'decoy receptor' that binds Gas6 with high affinity and inhibits its function, offering an alternative approach from drug discovery efforts that directly target Axl. Four mutations within this high-affinity Axl variant caused structural alterations in side chains across the Gas6-Axl binding interface, stabilizing a conformational change on Gas6. When reformatted as an Fc fusion, the engineered decoy receptor bound Gas6 with femtomolar affinity, an 80-fold improvement compared to binding of the wild-type Axl receptor, allowing effective sequestration of Gas6 and specific abrogation of Axl signaling. Moreover, increased Gas6 binding affinity was critical and correlative with the ability of decoy receptors to potently inhibit metastasis and disease progression in vivo.

Beyond antibodies: using biological principles to guide the development of next-generation protein therapeutics.

Protein-based biologics, which leverage the inherent affinity and specificity of protein-protein interactions, offer an effective strategy for targeting and modulating disease pathways. Despite the broad diversity of the proteome, monoclonal antibodies have been the major focus of such drug discovery efforts. While antibodies have shown great clinical value, the breadth and complexity of human disease highlight the need for alternatives that expand the therapeutic repertoire beyond this single class of proteins. The elucidation of molecular mechanisms underlying human disease has provided new opportunities for protein-based drugs to address challenging clinical problems. Natural ligands and receptors, which inherently modulate complex biological processes, have emerged as promising candidates for protein-based drug discovery efforts. Protein engineering strategies, guided by biological principles, are allowing ligands and receptors to be developed as next-generation therapeutics with improved safety and efficacy.

Role of a conserved arginine residue in linkers between the ketosynthase and acyltransferase domains of multimodular polyketide synthases.

The role of interdomain linkers in modular polyketide synthases is poorly understood. Analysis of the 6-deoxyerythronolide B synthase (DEBS) has yielded a model in which chain elongation is governed by interactions between the acyl carrier protein domain and the ketosynthase domain plus an adjacent linker. Alanine scanning mutagenesis of the conserved residues of this linker in DEBS module 3 led to the identification of the R513A mutant with a markedly reduced rate of chain elongation. Limited proteolysis supported a structural role for this Arg. Our findings highlight the importance of domain-linker interactions in assembly line polyketide biosynthesis.

Reprogramming a module of the 6-deoxyerythronolide B synthase for iterative chain elongation.

Multimodular polyketide synthases (PKSs) have an assembly line architecture in which a set of protein domains, known as a module, participates in one round of polyketide chain elongation and associated chemical modifications, after which the growing chain is translocated to the next PKS module. The ability to rationally reprogram these assembly lines to enable efficient synthesis of new polyketide antibiotics has been a long-standing goal in natural products biosynthesis. We have identified a ratchet mechanism that can explain the observed unidirectional translocation of the growing polyketide chain along the 6-deoxyerythronolide B synthase. As a test of this model, module 3 of the 6-deoxyerythronolide B synthase has been reengineered to catalyze two successive rounds of chain elongation. Our results suggest that high selectivity has been evolutionarily programmed at three types of protein-protein interfaces that are present repetitively along naturally occurring PKS assembly lines.

Probing the interactions of an acyl carrier protein domain from the 6-deoxyerythronolide B synthase.

The assembly-line architecture of polyketide synthases (PKSs) provides an opportunity to rationally reprogram polyketide biosynthetic pathways to produce novel antibiotics. A fundamental challenge toward this goal is to identify the factors that control the unidirectional channeling of reactive biosynthetic intermediates through these enzymatic assembly lines. Within the catalytic cycle of every PKS module, the acyl carrier protein (ACP) first collaborates with the ketosynthase (KS) domain of the paired subunit in its own homodimeric module so as to elongate the growing polyketide chain and then with the KS domain of the next module to translocate the newly elongated polyketide chain. Using NMR spectroscopy, we investigated the features of a structurally characterized ACP domain of the 6-deoxyerythronolide B synthase that contribute to its association with its KS translocation partner. Not only were we able to visualize selective protein-protein interactions between the two partners, but also we detected a significant influence of the acyl chain substrate on this interaction. A novel reagent, CF3-S-ACP, was developed as a ¹9F NMR spectroscopic probe of protein-protein interactions. The implications of our findings for understanding intermodular chain translocation are discussed.

Molecular recognition between ketosynthase and acyl carrier protein domains of the 6-deoxyerythronolide B synthase.

Every polyketide synthase module has an acyl carrier protein (ACP) and a ketosynthase (KS) domain that collaborate to catalyze chain elongation. The same ACP then engages the KS domain of the next module to facilitate chain transfer. Understanding the mechanism for this orderly progress of the growing polyketide chain represents a fundamental challenge in assembly line enzymology. Using both experimental and computational approaches, the molecular basis for KS-ACP interactions in the 6-deoxyerythronolide B synthase has been decoded. Surprisingly, KS-ACP recognition is controlled at different interfaces during chain elongation versus chain transfer. In fact, chain elongation is controlled at a docking site remote from the catalytic center. Not only do our findings reveal a new principle in the modular control of polyketide antibiotic biosynthesis, they also provide a rationale for the mandatory homodimeric structure of polyketide synthases, in contrast to the monomeric nonribosomal peptide synthetases.

Revisiting the modularity of modular polyketide synthases.

Modularity is a highly sought after feature in engineering design. A modular catalyst is a multi-component system whose parts can be predictably interchanged for functional flexibility and variety. Nearly two decades after the discovery of the first modular polyketide synthase (PKS), we critically assess PKS modularity in the face of a growing body of atomic structural and in vitro biochemical investigations. Both the architectural modularity and the functional modularity of this family of enzymatic assembly lines are reviewed, and the fundamental challenges that lie ahead for the rational exploitation of their full biosynthetic potential are discussed.

Mechanism based protein crosslinking of domains from the 6-deoxyerythronolide B synthase.

The critical role of protein-protein interactions in the chemistry of polyketide synthases is well established. However, the transient and weak nature of these interactions, in particular those involving the acyl carrier protein (ACP), has hindered efforts to structurally characterize these interactions. We describe a chemo-enzymatic approach that crosslinks the active sites of ACP and their cognate ketosynthase (KS) domains, resulting in the formation of a stable covalent adduct. This process is driven by specific protein-protein interactions between KS and ACP domains. Suitable manipulation of the reaction conditions enabled complete crosslinking of a representative KS and ACP, allowing isolation of a stable, conformationally constrained adduct suitable for high-resolution structural analysis.