Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data.

MOTIVATION: Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity, and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries. RESULTS: Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns. AVAILABILITY AND IMPLEMENTATION: All deep sequencing datasets and code to perform the analyses presented within are available via https://github.com/Tessier-Lab-UMich/PSERM_paper.

Mutational analysis of SARS-CoV-2 variants of concern reveals key tradeoffs between receptor affinity and antibody escape.

SARS-CoV-2 variants with enhanced transmissibility represent a serious threat to global health. Here we report machine learning models that can predict the impact of receptor-binding domain (RBD) mutations on receptor (ACE2) affinity, which is linked to infectivity, and escape from human serum antibodies, which is linked to viral neutralization. Importantly, the models predict many of the known impacts of RBD mutations in current and former Variants of Concern on receptor affinity and antibody escape as well as novel sets of mutations that strongly modulate both properties. Moreover, these models reveal key opposing impacts of RBD mutations on transmissibility, as many sets of RBD mutations predicted to increase antibody escape are also predicted to reduce receptor affinity and vice versa. These models, when used in concert, capture the complex impacts of SARS-CoV-2 mutations on properties linked to transmissibility and are expected to improve the development of next-generation vaccines and biotherapeutics.

Reduction of Glut1 in the Neural Retina But Not the RPE Alleviates Polyol Accumulation and Normalizes Early Characteristics of Diabetic Retinopathy.

Hyperglycemia is a key determinant for development of diabetic retinopathy (DR). Inadequate glycemic control exacerbates retinopathy, while normalization of glucose levels delays its progression. In hyperglycemia, hexokinase is saturated and excess glucose is metabolized to sorbitol by aldose reductase via the polyol pathway. Therapies to reduce retinal polyol accumulation for the prevention of DR have been elusive because of low sorbitol dehydrogenase levels in the retina and inadequate inhibition of aldose reductase. Using systemic and conditional genetic inactivation, we targeted the primary facilitative glucose transporter in the retina, Glut1, as a preventative therapeutic in diabetic male and female mice. Unlike WT diabetics, diabetic Glut1+/- mice did not display elevated Glut1 levels in the retina. Furthermore, diabetic Glut1+/- mice exhibited ameliorated ERG defects, inflammation, and oxidative stress, which was correlated with a significant reduction in retinal sorbitol accumulation. Retinal pigment epithelium-specific reduction of Glut1 did not prevent an increase in retinal sorbitol content or early hallmarks of DR. However, like diabetic Glut1+/- mice, reduction of Glut1 specifically in the retina mitigated polyol accumulation and diminished retinal dysfunction and the elevation of markers for oxidative stress and inflammation associated with diabetes. These results suggest that modulation of retinal polyol accumulation via Glut1 in photoreceptors can circumvent the difficulties in regulating systemic glucose metabolism and be exploited to prevent DR.SIGNIFICANCE STATEMENT Diabetic retinopathy affects one-third of diabetic patients and is the primary cause of vision loss in adults 20-74 years of age. While anti-VEGF and photocoagulation treatments for the late-stage vision threatening complications can prevent vision loss, a significant proportion of patients do not respond to anti-VEGF therapies, and mechanisms to stop progression of early-stage symptoms remain elusive. Glut1 is the primary facilitative glucose transporter for the retina. We determined that a moderate reduction in Glut1 levels, specifically in the retina, but not the retinal pigment epithelium, was sufficient to prevent retinal polyol accumulation and the earliest functional defects to be identified in the diabetic retina. Our study defines modulation of Glut1 in retinal neurons as a targetable molecule for prevention of diabetic retinopathy.

Rational affinity maturation of anti-amyloid antibodies with high conformational and sequence specificity.

The aggregation of amyloidogenic polypeptides is strongly linked to several neurodegenerative disorders, including Alzheimer's and Parkinson's diseases. Conformational antibodies that selectively recognize protein aggregates are leading therapeutic agents for selectively neutralizing toxic aggregates, diagnostic and imaging agents for detecting disease, and biomedical reagents for elucidating disease mechanisms. Despite their importance, it is challenging to generate high-quality conformational antibodies in a systematic and site-specific manner due to the properties of protein aggregates (hydrophobic, multivalent, and heterogeneous) and limitations of immunization (uncontrolled antigen presentation and immunodominant epitopes). Toward addressing these challenges, we have developed a systematic directed evolution procedure for affinity maturing antibodies against Alzheimer's Aß fibrils and selecting variants with strict conformational and sequence specificity. We first designed a library based on a lead conformational antibody by sampling combinations of amino acids in the antigen-binding site predicted to mediate high antibody specificity. Next, we displayed this library on the surface of yeast, sorted it against Aß42 aggregates, and identified promising clones using deep sequencing. The resulting antibodies displayed similar or higher affinities than clinical-stage Aß antibodies (aducanumab and crenezumab). Moreover, the affinity-matured antibodies retained high conformational specificity for Aß aggregates, as observed for aducanumab and unlike crenezumab. Notably, the affinity-maturated antibodies displayed extremely low levels of nonspecific interactions, as observed for crenezumab and unlike aducanumab. We expect that our systematic methods for generating antibodies with unique combinations of desirable properties will improve the generation of high-quality conformational antibodies specific for diverse types of aggregated conformers.

Antibodies with Weakly Basic Isoelectric Points Minimize Trade-offs between Formulation and Physiological Colloidal Properties.

The widespread interest in antibody therapeutics has led to much focus on identifying antibody candidates with favorable developability properties. In particular, there is broad interest in identifying antibody candidates with highly repulsive self-interactions in standard formulations (e.g., low ionic strength buffers at pH 5-6) for high solubility and low viscosity. Likewise, there is also broad interest in identifying antibody candidates with low levels of non-specific interactions in physiological solution conditions (PBS, pH 7.4) to promote favorable pharmacokinetic properties. To what extent antibodies that possess both highly repulsive self-interactions in standard formulations and weak non-specific interactions in physiological solution conditions can be systematically identified remains unclear and is a potential impediment to successful therapeutic drug development. Here, we evaluate these two properties for 42 IgG1 variants based on the variable fragments (Fvs) from four clinical-stage antibodies and complementarity-determining regions from 10 clinical-stage antibodies. Interestingly, we find that antibodies with the strongest repulsive self-interactions in a standard formulation (pH 6 and 10 mM histidine) display the strongest non-specific interactions in physiological solution conditions. Conversely, antibodies with the weakest non-specific interactions under physiological conditions display the least repulsive self-interactions in standard formulations. This behavior can be largely explained by the antibody isoelectric point, as highly basic antibodies that are highly positively charged under standard formulation conditions (pH 5-6) promote repulsive self-interactions that mediate high colloidal stability but also mediate strong non-specific interactions with negatively charged biomolecules at physiological pH and vice versa for antibodies with negatively charged Fv regions. Therefore, IgG1s with weakly basic isoelectric points between 8 and 8.5 and Fv isoelectric points between 7.5 and 9 typically display the best combinations of strong repulsive self-interactions and weak non-specific interactions. We expect that these findings will improve the identification and engineering of antibody candidates with drug-like biophysical properties.

Directed evolution of conformation-specific antibodies for sensitive detection of polypeptide aggregates in therapeutic drug formulations.

Biologics such as peptides and proteins possess a number of attractive attributes that make them particularly valuable as therapeutics, including their high activity, high specificity, and low toxicity. However, one of the key challenges associated with this class of drugs is their propensity to aggregate. Given the safety and immunogenicity concerns related to polypeptide aggregates, it is particularly important to sensitively detect aggregates in therapeutic drug formulations as part of the quality control process. Here, we report the development of conformation-specific antibodies that recognize polypeptide aggregates composed of a GLP-1 receptor agonist (liraglutide) and their integration into a sensitive immunoassay for detecting liraglutide amyloid fibrils. We sorted single-chain antibody libraries against liraglutide fibrils using yeast surface display and magnetic-activated cell sorting, and identified several antibodies with high conformational specificity. Interestingly, these antibodies cross-react with amyloid fibrils formed by several other polypeptides, revealing that they recognize molecular features common to different types of fibrils. Moreover, we find that our immunoassay using these antibodies is >50-fold more sensitive than the conventional method for detecting liraglutide aggregation (Thioflavin T fluorescence). We expect that our systematic approach for generating a sensitive, aggregate-specific immunoassay can be readily extended to other biologics to improve the quality and safety of formulated drug products.

Ultradilute Measurements of Self-Association for the Identification of Antibodies with Favorable High-Concentration Solution Properties.

There is significant interest in formulating antibody therapeutics as concentrated liquid solutions, but early identification of developable antibodies with optimal manufacturability, stability, and delivery attributes remains challenging. Traditional methods of identifying developable mAbs with low self-association in common antibody formulations require relatively concentrated protein solutions (>1 mg/mL), and this single challenge has frustrated early-stage and large-scale identification of antibody candidates with drug-like colloidal properties. Here, we describe charge-stabilized self-interaction nanoparticle spectroscopy (CS-SINS), an affinity-capture nanoparticle assay that measures colloidal self-interactions at ultradilute antibody concentrations (0.01 mg/mL), and is predictive of antibody developability issues of high viscosity and opalescence that manifest at four orders of magnitude higher concentrations (>100 mg/mL). CS-SINS enables large-scale, high-throughput selection of developable antibodies during early discovery.

Endogenous insulin signaling in the RPE contributes to the maintenance of rod photoreceptor function in diabetes.

In diabetes, there are two major physiological aberrations: (i) Loss of insulin signaling due to absence of insulin (type 1 diabetes) or insulin resistance (type 2 diabetes) and (ii) increased blood glucose levels. The retina has a high proclivity to damage following diabetes, and much of the pathology seen in diabetic retinopathy has been ascribed to hyperglycemia and downstream cascades activated by increased blood glucose. However, less attention has been focused on the direct role of insulin on retinal physiology, likely due to the fact that uptake of glucose in retinal cells is not insulin-dependent. The retinal pigment epithelium (RPE) is instrumental in maintaining the structural and functional integrity of the retina. Recent studies have suggested that RPE dysfunction is a precursor of, and contributes to, the development of diabetic retinopathy. To evaluate the role of insulin on RPE cell function directly, we generated a RPE specific insulin receptor (IR) knockout (RPEIRKO) mouse using the Cre-loxP system. Using this mouse, we sought to determine the impact of insulin-mediated signaling in the RPE on retinal function under physiological control conditions as well as in streptozotocin (STZ)-induced diabetes. We demonstrate that loss of RPE-specific IR expression resulted in lower a- and b-wave electroretinogram amplitudes in diabetic mice as compared to diabetic mice that expressed IR on the RPE. Interestingly, RPEIRKO mice did not exhibit significant differences in the amplitude of the RPE-dependent electroretinogram c-wave as compared to diabetic controls. However, loss of IR-mediated signaling in the RPE reduced levels of reactive oxygen species and the expression of pro-inflammatory cytokines in the retina of diabetic mice. These results imply that IR-mediated signaling in the RPE regulates photoreceptor function and may play a role in the generation of oxidative stress and inflammation in the retina in diabetes.

Optimization of therapeutic antibodies for reduced self-association and non-specific binding via interpretable machine learning.

Antibody development, delivery, and efficacy are influenced by antibody-antigen affinity interactions, off-target interactions that reduce antibody bioavailability and pharmacokinetics, and repulsive self-interactions that increase the stability of concentrated antibody formulations and reduce their corresponding viscosity. Yet identifying antibody variants with optimal combinations of these three types of interactions is challenging. Here we show that interpretable machine-learning classifiers, leveraging antibody structural features descriptive of their variable regions and trained on experimental data for a panel of 80 clinical-stage monoclonal antibodies, can identify antibodies with optimal combinations of low off-target binding in a common physiological-solution condition and low self-association in a common antibody-formulation condition. For three clinical-stage antibodies with suboptimal combinations of off-target binding and self-association, the classifiers predicted variable-region mutations that optimized non-affinity interactions while maintaining high-affinity antibody-antigen interactions. Interpretable machine-learning models may facilitate the optimization of antibody candidates for therapeutic applications.

Computational optimization of antibody humanness and stability by systematic energy-based ranking.

Conventional methods for humanizing animal-derived antibodies involve grafting their complementarity-determining regions onto homologous human framework regions. However, this process can substantially lower antibody stability and antigen-binding affinity, and requires iterative mutational fine-tuning to recover the original antibody properties. Here we report a computational method for the systematic grafting of animal complementarity-determining regions onto thousands of human frameworks. The method, which we named CUMAb (for computational human antibody design; available at http://CUMAb.weizmann.ac.il ), starts from an experimental or model antibody structure and uses Rosetta atomistic simulations to select designs by energy and structural integrity. CUMAb-designed humanized versions of five antibodies exhibited similar affinities to those of the parental animal antibodies, with some designs showing marked improvement in stability. We also show that (1) non-homologous frameworks are often preferred to highest-homology frameworks, and (2) several CUMAb designs that differ by dozens of mutations and that use different human frameworks are functionally equivalent.

Reduction of monoclonal antibody viscosity using interpretable machine learning.

Early identification of antibody candidates with drug-like properties is essential for simplifying the development of safe and effective antibody therapeutics. For subcutaneous administration, it is important to identify candidates with low self-association to enable their formulation at high concentration while maintaining low viscosity, opalescence, and aggregation. Here, we report an interpretable machine learning model for predicting antibody (IgG1) variants with low viscosity using only the sequences of their variable (Fv) regions. Our model was trained on antibody viscosity data (>100 mg/mL mAb concentration) obtained at a common formulation pH (pH 5.2), and it identifies three key Fv features of antibodies linked to viscosity, namely their isoelectric points, hydrophobic patch sizes, and numbers of negatively charged patches. Of the three features, most predicted antibodies at risk for high viscosity, including antibodies with diverse antibody germlines in our study (79 mAbs) as well as clinical-stage IgG1s (94 mAbs), are those with low Fv isoelectric points (Fv pIs < 6.3). Our model identifies viscous antibodies with relatively high accuracy not only in our training and test sets, but also for previously reported data. Importantly, we show that the interpretable nature of the model enables the design of mutations that significantly reduce antibody viscosity, which we confirmed experimentally. We expect that this approach can be readily integrated into the drug development process to reduce the need for experimental viscosity screening and improve the identification of antibody candidates with drug-like properties.

Simplifying complex antibody engineering using machine learning.

Machine learning is transforming antibody engineering by enabling the generation of drug-like monoclonal antibodies with unprecedented efficiency. Unsupervised algorithms trained on massive and diverse protein sequence datasets facilitate the prediction of panels of antibody variants with native-like intrinsic properties (e.g., high stability), greatly reducing the amount of subsequent experimentation needed to identify specific candidates that also possess desired extrinsic properties (e.g., high affinity). Additionally, supervised algorithms, which are trained on deep sequencing datasets obtained after enrichment of in vitro antibody libraries for one or more specific extrinsic properties, enable the prediction of antibody variants with desired combinations of extrinsic properties without the need for additional screening. Here we review recent advances using both machine learning approaches and how they are impacting the field of antibody engineering as well as key outstanding challenges and opportunities for these paradigm-changing methods.

Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data.

Motivation: Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries. Results: Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns. Availability: All deep sequencing datasets and code to do the analyses presented within are available via GitHub. Contact: Peter Tessier, ptessier@umich.edu. Supplementary information: Supplementary data are available at Bioinformatics online.

Quantitative flow cytometric selection of tau conformational nanobodies specific for pathological aggregates.

Single-domain antibodies, also known as nanobodies, are broadly important for studying the structure and conformational states of several classes of proteins, including membrane proteins, enzymes, and amyloidogenic proteins. Conformational nanobodies specific for aggregated conformations of amyloidogenic proteins are particularly needed to better target and study aggregates associated with a growing class of associated diseases, especially neurodegenerative disorders such as Alzheimer's and Parkinson's diseases. However, there are few reported nanobodies with both conformational and sequence specificity for amyloid aggregates, especially for large and complex proteins such as the tau protein associated with Alzheimer's disease, due to difficulties in selecting nanobodies that bind to complex aggregated proteins. Here, we report the selection of conformational nanobodies that selectively recognize aggregated (fibrillar) tau relative to soluble (monomeric) tau. Notably, we demonstrate that these nanobodies can be directly isolated from immune libraries using quantitative flow cytometric sorting of yeast-displayed libraries against tau aggregates conjugated to quantum dots, and this process eliminates the need for secondary nanobody screening. The isolated nanobodies demonstrate conformational specificity for tau aggregates in brain samples from both a transgenic mouse model and human tauopathies. We expect that our facile approach will be broadly useful for isolating conformational nanobodies against diverse amyloid aggregates and other complex antigens.

Flow cytometric isolation of drug-like conformational antibodies specific for amyloid fibrils.

Antibodies that recognize specific protein conformational states are broadly important for research, diagnostic and therapeutic applications, yet they are difficult to generate in a predictable and systematic manner using either immunization or in vitro antibody display methods. This problem is particularly severe for conformational antibodies that recognize insoluble antigens such as amyloid fibrils associated with many neurodegenerative disorders. Here we report a quantitative fluorescence-activated cell sorting (FACS) method for directly selecting high-quality conformational antibodies against different types of insoluble (amyloid fibril) antigens using a single, off-the-shelf human library. Our approach uses quantum dots functionalized with antibodies to capture insoluble antigens, and the resulting quantum dot conjugates are used in a similar manner as conventional soluble antigens for multi-parameter FACS selections. Notably, we find that this approach is robust for isolating high-quality conformational antibodies against tau and α-synuclein fibrils from the same human library with combinations of high affinity, high conformational specificity and, in some cases, low off-target binding that rival or exceed those of clinical-stage antibodies specific for tau (zagotenemab) and α-synuclein (cinpanemab). This approach is expected to enable conformational antibody selection and engineering against diverse types of protein aggregates and other insoluble antigens (e.g., membrane proteins) that are compatible with presentation on the surface of antibody-functionalized quantum dots.

Improving antibody drug development using bionanotechnology.

Monoclonal antibodies are being used to treat a remarkable breadth of human disorders. Nevertheless, there are several key challenges at the earliest stages of antibody drug development that need to be addressed using simple and widely accessible methods, especially related to generating antibodies against membrane proteins and identifying antibody candidates with drug-like biophysical properties (high solubility and low viscosity). Here we highlight key bionanotechnologies for preparing functional and stable membrane proteins in diverse types of lipoparticles that are being used to improve antibody discovery and engineering efforts. We also highlight key bionanotechnologies for high-throughput and ultra-dilute screening of antibody biophysical properties during antibody discovery and optimization that are being used for identifying antibodies with superior combinations of in vitro (formulation) and in vivo (half-life) properties.

Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space.

Therapeutic antibody development requires selection and engineering of molecules with high affinity and other drug-like biophysical properties. Co-optimization of multiple antibody properties remains a difficult and time-consuming process that impedes drug development. Here we evaluate the use of machine learning to simplify antibody co-optimization for a clinical-stage antibody (emibetuzumab) that displays high levels of both on-target (antigen) and off-target (non-specific) binding. We mutate sites in the antibody complementarity-determining regions, sort the antibody libraries for high and low levels of affinity and non-specific binding, and deep sequence the enriched libraries. Interestingly, machine learning models trained on datasets with binary labels enable predictions of continuous metrics that are strongly correlated with antibody affinity and non-specific binding. These models illustrate strong tradeoffs between these two properties, as increases in affinity along the co-optimal (Pareto) frontier require progressive reductions in specificity. Notably, models trained with deep learning features enable prediction of novel antibody mutations that co-optimize affinity and specificity beyond what is possible for the original antibody library. These findings demonstrate the power of machine learning models to greatly expand the exploration of novel antibody sequence space and accelerate the development of highly potent, drug-like antibodies.

Reduction of therapeutic antibody self-association using yeast-display selections and machine learning.

Self-association governs the viscosity and solubility of therapeutic antibodies in high-concentration formulations used for subcutaneous delivery, yet it is difficult to reliably identify candidates with low self-association during antibody discovery and early-stage optimization. Here, we report a high-throughput protein engineering method for rapidly identifying antibody candidates with both low self-association and high affinity. We find that conjugating quantum dots to IgGs that strongly self-associate (pH 7.4, PBS), such as lenzilumab and bococizumab, results in immunoconjugates that are highly sensitive for detecting other high self-association antibodies. Moreover, these conjugates can be used to rapidly enrich yeast-displayed bococizumab sub-libraries for variants with low levels of immunoconjugate binding. Deep sequencing and machine learning analysis of the enriched bococizumab libraries, along with similar library analysis for antibody affinity, enabled identification of extremely rare variants with co-optimized levels of low self-association and high affinity. This analysis revealed that co-optimizing bococizumab is difficult because most high-affinity variants possess positively charged variable domains and most low self-association variants possess negatively charged variable domains. Moreover, negatively charged mutations in the heavy chain CDR2 of bococizumab, adjacent to its paratope, were effective at reducing self-association without reducing affinity. Interestingly, most of the bococizumab variants with reduced self-association also displayed improved folding stability and reduced nonspecific binding, revealing that this approach may be particularly useful for identifying antibody candidates with attractive combinations of drug-like properties.Abbreviations: AC-SINS: affinity-capture self-interaction nanoparticle spectroscopy; CDR: complementarity-determining region; CS-SINS: charge-stabilized self-interaction nanoparticle spectroscopy; FACS: fluorescence-activated cell sorting; Fab: fragment antigen binding; Fv: fragment variable; IgG: immunoglobulin; QD: quantum dot; PBS: phosphate-buffered saline; VH: variable heavy; VL: variable light.

Discovery-stage identification of drug-like antibodies using emerging experimental and computational methods.

There is intense and widespread interest in developing monoclonal antibodies as therapeutic agents to treat diverse human disorders. During early-stage antibody discovery, hundreds to thousands of lead candidates are identified, and those that lack optimal physical and chemical properties must be deselected as early as possible to avoid problems later in drug development. It is particularly challenging to characterize such properties for large numbers of candidates with the low antibody quantities, concentrations, and purities that are available at the discovery stage, and to predict concentrated antibody properties (e.g., solubility, viscosity) required for efficient formulation, delivery, and efficacy. Here we review key recent advances in developing and implementing high-throughput methods for identifying antibodies with desirable in vitro and in vivo properties, including favorable antibody stability, specificity, solubility, pharmacokinetics, and immunogenicity profiles, that together encompass overall drug developability. In particular, we highlight impressive recent progress in developing computational methods for improving rational antibody design and prediction of drug-like behaviors that hold great promise for reducing the amount of required experimentation. We also discuss outstanding challenges that will need to be addressed in the future to fully realize the great potential of using such analysis for minimizing development times and improving the success rate of antibody candidates in the clinic.