Prediction of delayed retention of antibodies in hydrophobic interaction chromatography from sequence using machine learning.

MOTIVATION: The hydrophobicity of a monoclonal antibody is an important biophysical property relevant for its developability into a therapeutic. In addition to characterizing heterogeneity, Hydrophobic Interaction Chromatography (HIC) is an assay that is often used to quantify the hydrophobicity of an antibody to assess downstream risks. Earlier studies have shown that retention times in this assay can be correlated to amino-acid or atomic propensities weighted by the surface areas obtained from protein 3-dimensional structures. The goal of this study is to develop models to enable prediction of delayed HIC retention times directly from sequence. RESULTS: We utilize the randomforest machine learning approach to estimate the surface exposure of amino-acid side-chains in the variable region directly from the antibody sequence. We obtain mean-absolute errors of 4.6% for the prediction of surface exposure. Using experimental HIC data along with the estimated surface areas, we derive an amino-acid propensity scale that enables prediction of antibodies likely to have delayed retention times in the assay. We achieve a cross-validation Area Under Curve of 0.85 for the Receiver Operating Characteristic curve of our model. The low computational expense and high accuracy of this approach enables real-time assessment of hydrophobic character to enable prioritization of antibodies during the discovery process and rational engineering to reduce hydrophobic liabilities. AVAILABILITY AND IMPLEMENTATION: Structure data, aligned sequences, experimental data and prediction scores for test-cases, and R scripts used in this work are provided as part of the Supplementary Material. CONTACT: tushar.jain@adimab.com. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Production of sialylated O-linked glycans in Pichia pastoris.

The methylotrophic yeast, Pichia pastoris, is an important organism used for the production of therapeutic proteins. Previously, we have reported the glycoengineering of this organism to produce human-like N-linked glycans but up to now no one has addressed engineering the O-linked glycosylation pathway. Typically, O-linked glycans produced by wild-type P. pastoris are linear chains of four to five α-linked mannose residues, which may be capped with ß- or phospho-mannose. Previous genetic engineering of the N-linked glycosylation pathway of P. pastoris has eliminated both of these two latter modifications, resulting in O-linked glycans which are linear α-linked mannose structures. Here, we describe a method for the co-expression of an α-1,2-mannosidase, which reduces these glycans to primarily a single O-linked mannose residue. In doing so, we have reduced the potential of these glycans to interact with carbohydrate-binding proteins, such as dendritic cell-specific intercellular adhesion molecule-3-grabbing non-integrin. Furthermore, the introduction of the enzyme protein-O-linked-mannose ß-1,2-N-acetylglucosaminyltransferase 1, resulted in the capping of the single O-linked mannose residues with N-acetylglucosamine. Subsequently, this glycoform was extended into human-like sialylated glycans, similar in structure to α-dystroglycan-type glycoforms. As such, this represents the first example of sialylated O-linked glycans being produced in yeast and extends the utility of the P. pastoris production platform beyond N-linked glycosylated biotherapeutics to include molecules possessing O-linked glycans.

N-glycosylamine-mediated isotope labeling for mass spectrometry-based quantitative analysis of N-linked glycans.

N-linked glycosylation is a major protein modification involved in many essential cellular functions. Methods capable of quantitative glycan analysis are highly valuable and have been actively pursued. Here we describe a novel N-glycosylamine-based strategy for isotopic labeling of N-linked glycans for quantitative analysis by use of mass spectrometry (MS). This strategy relies on the primary amine group on the reducing end of freshly released N-linked glycans for labeling, and eliminates the need for the harsh labeling reaction conditions and/or tedious cleanup procedures required by existing methods. By using NHS-ester amine chemistry we used this strategy to label N-linked glycans from a monoclonal antibody with commercially available tandem mass tags (TMT). Only duplex experiments can be performed with currently available TMT reagents, because quantification is based on the intensity of intact labeled glycans. Under mild reaction conditions, greater than 95% derivatization was achieved in 30 min and the labeled glycans, when kept at -20 °C, were stable for more than 10 days. By performing glycan release, TMT labeling, and LC-MS analysis continuously in a single volatile aqueous buffer without cleanup steps, we were able to complete the entire analysis in less than 2 h. Quantification was highly accurate and the dynamic range was large. Compared with previously established methods, N-glycosylamine-mediated labeling has the advantages of experimental simplicity, efficient labeling, and preserving glycan integrity.

High-throughput screening and selection of yeast cell lines expressing monoclonal antibodies.

The methylotrophic yeast Pichia pastoris has recently been engineered to express therapeutic glycoproteins with uniform human N-glycans at high titers. In contrast to the current art where producing therapeutic proteins in mammalian cell lines yields a final product with heterogeneous N-glycans, proteins expressed in glycoengineered P. pastoris can be designed to carry a specific, preselected glycoform. However, significant variability exists in fermentation performance between genotypically similar clones with respect to cell fitness, secreted protein titer, and glycan homogeneity. Here, we describe a novel, multidimensional screening process that combines high and medium throughput tools to identify cell lines producing monoclonal antibodies (mAbs). These cell lines must satisfy multiple selection criteria (high titer, uniform N-glycans and cell robustness) and be compatible with our large-scale production platform process. Using this selection process, we were able to isolate a mAb-expressing strain yielding a titer (after protein A purification) in excess of 1 g/l in 0.5-l bioreactors.

Assessing the binding properties of the anti-PD-1 antibody landscape using label-free biosensors.

Here we describe an industry-wide collaboration aimed at assessing the binding properties of a comprehensive panel of monoclonal antibodies (mAbs) against programmed cell death protein 1 (PD-1), an important checkpoint protein in cancer immunotherapy and validated therapeutic target, with well over thirty unique mAbs either in clinical development or market-approved in the United States, the European Union or China. The binding kinetics of the PD-1/mAb interactions were measured by surface plasmon resonance (SPR) using a Carterra LSA instrument and the results were compared to data collected on a Biacore 8K. The effect of chip type on the SPR-derived binding rate constants and affinities were explored and the results compared with solution affinities from Meso Scale Discovery (MSD) and Kinetic Exclusion Assay (KinExA) experiments. When using flat chip types, the LSA and 8K platforms yielded near-identical kinetic rate and affinity constants that matched solution phase values more closely than those produced on 3D-hydrogels. Of the anti-PD-1 mAbs tested, which included a portion of those known to be in clinical development or approved, the affinities spanned from single digit picomolar to nearly 425 nM, challenging the dynamic range of our methods. The LSA instrument was also used to perform epitope binning and ligand competition studies which revealed over ten unique competitive binding profiles within this group of mAbs.

Broad sarbecovirus neutralizing antibodies define a key site of vulnerability on the SARS-CoV-2 spike protein.

Broadly protective vaccines against known and pre-emergent coronaviruses are urgently needed. Critical to their development is a deeper understanding of cross-neutralizing antibody responses induced by natural human coronavirus (HCoV) infections. Here, we mined the memory B cell repertoire of a convalescent SARS donor and identified 200 SARS-CoV-2 binding antibodies that target multiple conserved sites on the spike (S) protein. A large proportion of the antibodies display high levels of somatic hypermutation and cross-react with circulating HCoVs, suggesting recall of pre-existing memory B cells (MBCs) elicited by prior HCoV infections. Several antibodies potently cross-neutralize SARS-CoV, SARS-CoV-2, and the bat SARS-like virus WIV1 by blocking receptor attachment and inducing S1 shedding. These antibodies represent promising candidates for therapeutic intervention and reveal a new target for the rational design of pan-sarbecovirus vaccines.

Broad neutralization of SARS-related viruses by human monoclonal antibodies.

Broadly protective vaccines against known and preemergent human coronaviruses (HCoVs) are urgently needed. To gain a deeper understanding of cross-neutralizing antibody responses, we mined the memory B cell repertoire of a convalescent severe acute respiratory syndrome (SARS) donor and identified 200 SARS coronavirus 2 (SARS-CoV-2) binding antibodies that target multiple conserved sites on the spike (S) protein. A large proportion of the non-neutralizing antibodies display high levels of somatic hypermutation and cross-react with circulating HCoVs, suggesting recall of preexisting memory B cells elicited by prior HCoV infections. Several antibodies potently cross-neutralize SARS-CoV, SARS-CoV-2, and the bat SARS-like virus WIV1 by blocking receptor attachment and inducing S1 shedding. These antibodies represent promising candidates for therapeutic intervention and reveal a target for the rational design of pan-sarbecovirus vaccines.

Affinity Maturation Enhances Antibody Specificity but Compromises Conformational Stability.

Monoclonal antibodies (mAbs) have recently emerged as one of the most promising classes of biotherapeutics. A potential advantage of B cell-derived mAbs as therapeutic agents is that they have been subjected to natural filtering mechanisms, which may enrich for B cell receptors (BCRs) with favorable biophysical properties. Here, we evaluated 400 human mAbs for polyreactivity, hydrophobicity, and thermal stability using high-throughput screening assays. Overall, mAbs derived from memory B cells and long-lived plasma cells (LLPCs) display reduced levels of polyreactivity, hydrophobicity, and thermal stability compared with naive B cell-derived mAbs. Somatic hypermutation (SHM) is inversely associated with all three biophysical properties, as well as BCR expression levels. Finally, the developability profiles of the human B cell-derived mAbs are comparable with those observed for clinical mAbs, suggesting their high therapeutic potential. The results provide insight into the biophysical consequences of affinity maturation and have implications for therapeutic antibody engineering and development.

Rapid assessment of oxidation via middle-down LCMS correlates with methionine side-chain solvent-accessible surface area for 121 clinical stage monoclonal antibodies.

Susceptibility of methionine to oxidation is an important concern for chemical stability during the development of a monoclonal antibody (mAb) therapeutic. To minimize downstream risks, leading candidates are usually screened under forced oxidation conditions to identify oxidation-labile molecules. Here we report results of forced oxidation on a large set of in-house expressed and purified mAbs with variable region sequences corresponding to 121 clinical stage mAbs. These mAb samples were treated with 0.1% H2O2 for 24 hours before enzymatic cleavage below the hinge, followed by reduction of inter-chain disulfide bonds for the detection of the light chain, Fab portion of heavy chain (Fd) and Fc by liquid chromatography-mass spectrometry. This high-throughput, middle-down approach allows detection of oxidation site(s) at the resolution of 3 distinct segments. The experimental oxidation data correlates well with theoretical predictions based on the solvent-accessible surface area of the methionine side-chains within these segments. These results validate the use of upstream computational modeling to predict mAb oxidation susceptibility at the sequence level.

Database-Centric Method for Automated High-Throughput Deconvolution and Analysis of Kinetic Antibody Screening Data.

The state-of-the-art industrial drug discovery approach is the empirical interrogation of a library of drug candidates against a target molecule. The advantage of high-throughput kinetic measurements over equilibrium assessments is the ability to measure each of the kinetic components of binding affinity. Although high-throughput capabilities have improved with advances in instrument hardware, three bottlenecks in data processing remain: (1) intrinsic molecular properties that lead to poor biophysical quality in vitro are not accounted for in commercially available analysis models, (2) processing data through a user interface is time-consuming and not amenable to parallelized data collection, and (3) a commercial solution that includes historical kinetic data in the analysis of kinetic competition data does not exist. Herein, we describe a generally applicable method for the automated analysis, storage, and retrieval of kinetic binding data. This analysis can deconvolve poor quality data on-the-fly and store and organize historical data in a queryable format for use in future analyses. Such database-centric strategies afford greater insight into the molecular mechanisms of kinetic competition, allowing for the rapid identification of allosteric effectors and the presentation of kinetic competition data in absolute terms of percent bound to antigen on the biosensor.

Understanding ForteBio's Sensors for High-Throughput Kinetic and Epitope Screening for Purified Antibodies and Yeast Culture Supernatant.

Real-time and label-free antibody screening systems are becoming more popular because of the increasing output of purified antibodies and antibody supernatant from many antibody discovery platforms. However, the properties of the biosensor can greatly affect the kinetic and epitope binning results generated by these label-free screening systems. ForteBio human-specific ProA, anti-human IgG quantitation (AHQ), anti-human Fc capture (AHC) sensors, and custom biotinylated-anti-human Fc capture (b-AHFc) sensors were evaluated in terms of loading ability, regeneration, kinetic characterization, and epitope binning with both purified IgG and IgG supernatant. AHC sensors proved unreliable for kinetic or binning assays at times, whereas AHQ sensors showed poor loading and regeneration abilities. ProA sensors worked well with both purified IgG and IgG supernatant. However, the interaction between ProA sensors and the Fab region of the IgG with VH3 germline limited the application of ProA sensors, especially in the epitope binning experiment. In an attempt to generate a biosensor type that would be compatible with a variety of germlines and sample types, we found that the custom b-AHFc sensors appeared to be robust working with both purified IgG and IgG supernatant, with little evidence of sensor-related artifacts.

O-linked glycosylation analysis of recombinant human granulocyte colony-stimulating factor produced in glycoengineered Pichia pastoris by liquid chromatography and mass spectrometry.

Glycosylation is a major biochemical attribute of therapeutic proteins and detailed analyses including the structures and sites of such modifications are often required for product quality control and assurance. Using liquid chromatography and tandem mass spectrometry techniques, we analyzed the O-linked glycosylation of recombinant human granulocyte colony-stimulating factor (rhG-CSF) derived from glycoengineered Pichia pastoris with regard to its nature, structure, occupancy, and location. Peptide mappings using protease and chemical cleavages were performed to determine the specific O-linked glycosylation site used by Pichia-derived rhG-CSF. Our results demonstrated that Thr134, the equivalent O-linked glycosylation site found on endogenous human G-CSF, is the only site modified with a single mannose, allowing glycoengineered P. pastoris to be used as a viable production platform for therapeutic rhG-CSF.

A cost-effective plate-based sample preparation for antibody N-glycan analysis.

During early cell line and process development of therapeutic antibodies, a cost-effective high-throughput approach to characterize the N-linked glycans is highly desired given that a large number of samples need to be analyzed. Using commercially available, low cost 96-well plates, we developed a practical procedure to prepare fluorescently labeled N-linked glycans for both qualitative and quantitative analysis by mass spectrometry (MS) and ultrahigh performance liquid chromatography (UPLC). Antibody samples were continuously denatured, reduced, and deglycosylated in a single 96-well hydrophobic membrane filter plate. Subsequently, released glycans were fluorescently labeled in a collection plate, and cleaned-up using a hydrophilic membrane filter plate. Carried out entirely in ready-to-use 96-well plates with simple buffer systems, this procedure requires less than 90min to finish. We applied the optimized procedure to examine the N-linked glycosylation of trastuzumab and were able to quantify ten major N-linked glycans. The results from different amounts of starting materials (10-200µg) were highly similar and showed the robustness of this procedure. Compared to other methods, this new procedure is simple to implement, economically more affordable, and could be very valuable for early screenings of antibody development.

Glycosylation characterization of recombinant human erythropoietin produced in glycoengineered Pichia pastoris by mass spectrometry.

Glycosylation plays a critical role in the in vivo efficacy of both endogenous and recombinant erythropoietin (EPO). Using mass spectrometry, we characterized the N-/O-linked glycosylation of recombinant human EPO (rhEPO) produced in glycoengineered Pichia pastoris and compared with the glycosylation of Chinese hamster ovary (CHO) cell-derived rhEPO. While the three predicted N-linked glycosylation sites (Asn24, Asn38 and Asn83) showed complete site occupancy, Pichia- and CHO-derived rhEPO showed distinct differences in the glycan structures with the former containing sialylated bi-antennary glycoforms and the latter containing a mixture of sialylated bi-, tri- and tetra-antennary structures. Additionally, the N-linked glycans from Pichia-produced rhEPO were similar across all three sites. A low level of O-linked mannosylation was detected on Pichia-produced rhEPO at position Ser126, which is also the O-linked glycosylation site for endogenous human EPO and CHO-derived rhEPO. In summary, the mass spectrometric analyses revealed that rhEPO derived from glycoengineered Pichia has a highly uniform bi-antennary N-linked glycan composition and preserves the orthogonal O-linked glycosylation site present on endogenous human EPO and CHO-derived rhEPO.

Characterization of the Pichia pastoris protein-O-mannosyltransferase gene family.

The methylotrophic yeast, Pichiapastoris, is an important organism used for the production of therapeutic proteins. However, the presence of fungal-like glycans, either N-linked or O-linked, can elicit an immune response or enable the expressed protein to bind to mannose receptors, thus reducing their efficacy. Previously we have reported the elimination of ß-linked glycans in this organism. In the current report we have focused on reducing the O-linked mannose content of proteins produced in P. pastoris, thereby reducing the potential to bind to mannose receptors. The initial step in the synthesis of O-linked glycans in P. pastoris is the transfer of mannose from dolichol-phosphomannose to a target protein in the yeast secretory pathway by members of the protein-O-mannosyltransferase (PMT) family. In this report we identify and characterize the members of the P. pastoris PMT family. Like Candida albicans, P. pastoris has five PMT genes. Based on sequence homology, these PMTs can be grouped into three sub-families, with both PMT1 and PMT2 sub-families possessing two members each (PMT1 and PMT5, and PMT2 and PMT6, respectively). The remaining sub-family, PMT4, has only one member (PMT4). Through gene knockouts we show that PMT1 and PMT2 each play a significant role in O-glycosylation. Both, by gene knockouts and the use of Pmt inhibitors we were able to significantly reduce not only the degree of O-mannosylation, but also the chain-length of these glycans. Taken together, this reduction of O-glycosylation represents an important step forward in developing the P. pastoris platform as a suitable system for the production of therapeutic glycoproteins.

Binding of DC-SIGN to glycoproteins expressed in glycoengineered Pichia pastoris.

Previous studies have shown that glycoproteins expressed in wild-type Pichia pastoris bind to Dendritic cell-SIGN (DC-Specific Intercellular adhesion molecule-3 Grabbing Nonintegrin), a mannose-binding receptor found on dendritic cells in peripheral tissues which is involved in antigen presentation and the initiation of an immune response. However, the binding of DC-SIGN to glycoproteins purified from P. pastoris strains engineered to express humanized N- and O-linked glycans has not been tested to date. In this study, the binding of glycoproteins with specific high-mannose or human N- and O-linked glycan structures to DC-SIGN was tested. Proteins with humanized N-glycans including Man5 structures and O-glycans (up to as many as 24) with single mannose chain length showed DC-SIGN binding that was comparable to that measured for a CHO-produced IgG1 which lacks O-linked mannose. Glycoproteins with wild-type N-glycans and mannotriose and higher O-glycans bound to DC-SIGN in a manner that was strongly inhibited by either the use of enzymatic N-deglycosylation or sodium meta-periodate oxidation. Mannan purified from humanized P. pastoris also showed lower ability to inhibit DC-SIGN binding to glycoproteins with wild type fungal glycosylation than mannan purified from wild type strains. This study shows that humanized P. pastoris can produce glycoproteins that do not bind to DC-SIGN.

Glycoengineered Pichia produced anti-HER2 is comparable to trastuzumab in preclinical study.

Mammalian cell culture systems are used predominantly for the production of therapeutic monoclonal antibody (mAb) products. A number of alternative platforms, such as Pichia engineered with a humanized N-linked glycosylation pathway, have recently been developed for the production of mAbs. The glycosylation profiles of mAbs produced in glycoengineered Pichia are similar to those of mAbs produced in mammalian systems. This report presents for the first time the comprehensive characterization of an anti-human epidermal growth factor receptor 2 (HER2) mAb produced in a glycoengineered Pichia, and a study comparing the anti-HER2 from Pichia, which had an amino acid sequence identical to trastuzumab, with trastuzumab. The comparative study covered a full spectrum of preclinical evaluation, including bioanalytical characterization, in vitro biological functions, in vivo anti-tumor efficacy and pharmacokinetics in both mice and non-human primates. Cell signaling and proliferation assays showed that anti-HER2 from Pichia had antagonist activities comparable to trastuzumab. However, Pichia-produced material showed a 5-fold increase in binding affinity to FcγIIIA and significantly enhanced antibody dependant cell-mediated cytotoxicity (ADCC) activity, presumably due to the lack of fucose on N-glycans. In a breast cancer xenograft mouse model, anti-HER2 was comparable to trastuzumab in tumor growth inhibition. Furthermore, comparable pharmacokinetic profiles were observed for anti-HER2 and trastuzumab in both mice and cynomolgus monkeys. We conclude that glycoengineered Pichia provides an alternative production platform for therapeutic mAbs and may be of particular interest for production of antibodies for which ADCC is part of the clinical mechanism of action.

Use of high-performance anion exchange chromatography with pulsed amperometric detection for O-glycan determination in yeast.

O-glycosylation is a post-translational protein modification that occurs in all eukaryotes. Yeasts have received increasing attention as a host for therapeutic protein production because of their ability to secrete high levels of recombinant protein. Because yeasts such as Pichia pastoris have been shown to O-glycosylate some proteins with varying effects on protein function, it is important to elucidate the nature of this modification. Methods that characterize O-glycosylation on a qualitative and quantitative basis are thus important when considering yeast as a host for therapeutic protein production. This protocol describes the release of O-glycans from a protein sample by -elimination under alkaline conditions using sodium borohydride and sodium hydroxide. The released O-linked oligosaccharides are subsequently processed and then separated by high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD). An estimation of O-glycan molar occupancy and average O-mannose chain length is ultimately derived. This protocol requires approximately 3 d for completion. This method provides an assessment of O-glycosylation and allows one to correlate the effect of O-glycosylation on protein properties.