Optimizing asymmetric antibody purification: a semi-automated process and its digital integration

ABSTRACT

Over the past decades, immunization and display technologies have considerably increased the potential for generating new binders against cell surface targets. Concomitantly, the complexity of biologic therapeutic drugs has also increased, with new asymmetric formats such as bispecific antibodies or antibody fusion proteins making the supply of molecules for preclinical drug discovery more challenging. The purification of those molecules is crucial, and an efficient purification platform for drug discovery research units should have multiple aims. First, it needs to deliver the highest quality proteins for research activities at a fast pace in order to increase screening capacities. Second, it has to deliver protein with sufficient yield in order to cover the project requirements and minimize the repetition of production cycles. Through a case study for a bispecific antibody, we describe a semi-automated and digitalized purification platform aiming at accelerating and optimizing the supply of asymmetric antibodies for drug discovery. We show how the automation of repetitive tasks and the digitalization of the process can lead to increased throughput in the context of complex purifications, including a cation exchange chromatography separation step. Furthermore, we highlight how process digitalization leads to enhanced data capture and accessibility, facilitating decision-making along the purification process. With a maximal throughput of 36 asymmetric antibodies per week and data proving the consistency of the quality delivered, this platform represents a step forward in the supply of complex antibody formats for preclinical drug discovery.

Introduction

Monoclonal antibodies have played a critical role in scientific research for decades, and numerous antibodies have become clinical treatments. Over the past 20 y, advancements in antibody engineering have enabled the development of innovative asymmetric antibody drug formats, such as bispecific antibodies (bsAbs), trispecific antibodies, or antibody fusion proteins. Unlike natural monospecific antibodies, multispecific antibodies have the ability to recognize different epitopes, either on the same or on different antigens. This feature enables binding to antigens with multiple domains, enhancing avidity and/or specificity, which results in superior activity in preclinical models compared to monovalent binding. By the end of 2024, 18 bsAbs had received approval as therapeutic drugs.^1–7

The bsAbs format has an impact on the mode of action, and therefore format standardization is unlikely. Nonetheless, recent approvals highlight a focus on classical heterodimeric IgG-like formats. These IgG-like molecules exhibit pharmacokinetics akin to conventional antibodies, attributed to the presence of an Fc region, and display favorable solubility and stability, rendering them excellent candidates for technical development and manufacturing. However, the production of bsAbs or antibody fusion proteins requires the combination of different heavy (HC) and light (LC) chains, leading to inherent asymmetry and the risk of mispairing of these chains. Ensuring the production of high-quality molecules and separating the desired antibodies from unwanted by-products remains a significant challenge. To address chain mispairing, several strategies have been developed. For instance, correct heavy chain association in asymmetric IgG-like bsAbs can be achieved using the knobs-into-holes (KiH) technology, as exemplified by faricimab, or through controlled Fab-arm exchange (cFAE), demonstrated by amivantamab.^1,6,8–10

The supply of asymmetric antibodies, here referred to as bsAbs and antibody fusion proteins, represents a challenge along the entire product lifecycle. Especially in the early discovery phase, the variety of drug candidates and the amount needed for in-depth characterization represent a substantial challenge for a streamlined generic supply process. Contemporary antibody selection techniques, including immunization associated with B-cell cloning or phage display screening of protein libraries, allow identification of thousands of binders against specific targets and homologues.¹¹ Following selection, the next step is the identification of paratope combinations and their optimal configuration to best achieve the desired mechanism of action when assembled as bispecific molecules. Different approaches have been described to explore the resulting design space, since the identification of suitable molecules for therapeutic drug development has been hindered by the complexity of supplying large numbers of binder-format combinations.¹²

In order to accelerate access to innovations for patients, both speed and cost-efficiency are major focuses of research and development organizations. The development of automated high throughput methods is crucial to increase functional screening capacities and reduce cycle times of therapeutic antibody discovery. High-throughput purification platforms based on 96-well plates and using liquid handling systems have been described.¹³ However, these methods do not provide sufficient quantities of asymmetric antibodies with the required purity for multiple assays and in depth characterization. Consequently, new approaches are required to enhance throughput while maintaining quality standards and scale.

A semi-automated process described by Yang et al.¹⁴ enables simultaneous affinity chromatography for up to 24 samples on a ProteinMaker instrument, followed by a preparative size exclusion chromatography (pSEC), and has proven effective for standard antibody purifications.^14–16 The purification of asymmetric antibodies traditionally requires the addition of a separation step between the affinity chromatography and the polishing pSEC step in order to separate the molecule of interest from product-related impurities. Cation exchange chromatography (CEX) is in most cases suitable for the separation of the side products.¹⁷ This additional chromatography step and the associated in-process analysis significantly increase the workload associated with the downstream processing of each molecule. Furthermore, the multiplication of tasks when purifying several molecules in parallel makes the processes tedious and extremely challenging. Creation of a platform that combines the necessary equipment for efficient purification processes (e.g., chromatography systems, analytical devices), advanced automation capabilities, and a unified interface for digital process orchestration and data management was thus a key objective. Critically, the use of automated liquid handling platforms accelerates laboratory processes and minimizes errors. Such platforms enable the execution of large experiments and the collection of extensive datasets with minimal operator intervention.¹⁸ The development of the Roche “AutoLab” automation platform provides an intuitive and unified user interface to fulfill laboratory operations. It orchestrates complex workflows and integrates the different components involved, such as laboratory instruments and databases. By doing so, it enables control of the instruments but also the efficient capture, review, and storage of process data in real time, taking automation and data management to the next level.^19,20

Our case study outlines the transfer of a standard three-step purification process to a semi-automated ProteinMaker-based platform controlled via Autolab, resulting in a significant increase in throughput. We describe the process, assess the relevance of the in-process control strategy, and evaluate the performance of the newly established purification platform.

Materials and methods

Software

In contemporary data management systems, user interface (UI) driven data capturing solutions are prevalent due to their initial simplicity and ease of implementation. However, these systems often encounter several bottlenecks that impede their long-term efficiency and scalability. To overcome these challenges, Autolab, an automation platform leveraging off-the-shelf solutions and custom components, has been developed. The key components of Autolab include Camunda, a robust workflow engine that orchestrates complex workflows ensuring efficient execution of business processes.²¹ MinIO, a high performance and scalable data storage system, is used for temporary data capture.²² Various interfaces are provided to integrate with laboratory equipment (e.g., RabbitMQ, RestFul API). To facilitate data transformation processes, a Python-based broker was developed to execute the necessary data transformations. The primary objective of this platform was to provide a unified framework for data acquisition and device integration. To enable user interaction with the platform, a web-based user interface was implemented. This interface allows users to configure and initiate device operations, as well as to capture and review analytical results (Figure 1).

Figure 1.

Overview of the three stages of the process orchestrated through the Roche Autolab platform from harvest to storage of the final product.

Molecule

The bispecific molecule (bsAb1) (Figure 2a) was assembled using the following polypeptide chains: A) VH-CL domain against target 1, K) crossed VL-CH domain against target 1 fused to huIgG1 Fc knob chain, H) VH-CH domain against target 2 containing charge mutations CH1(EE) fused to huIgG1 Fc hole chain, B) VL-CL domain against target 2 containing charge mutations CL(RK).²³ Additionally, the Pro329Gly, Leu234Ala and Leu235Ala mutations were introduced to abrogate binding to Fc receptors.²⁴ To ensure correct assembly and prevent mispairing, the knob into hole (KiH) technology was used.²⁵

Figure 2.

Description and quality assessment of the 1+1 IgG-like bsAb1 molecule. (a) Molecule chain assembly (ABHK), physicochemical properties and illustration of potential side products. (b) Results of aSEC and CE-SDS performed with the PA micro-purified material. The product peak % in the non-reduced CE-SDS electropherogram and the product peak SEC % in aSEC chromatogram define the area containing the target molecule bsAb1. The legend of the reduced CE-SDS electropherogram indicates the chain labels used to calculate the chain ratio in the results part. Figure 2a was created in BioRender. Georgiadou, D. (2025) https://BioRender.com/z50c989.

Expression

The bispecific antibody bsAb1 was produced in Chinese hamster ovary (CHO) cells by transient expression using a plasmid ratio based on the stoichiometry of the molecule (1:1:1:1).

Micropurification for expression quality assessment

To assess the correct pairing of the molecule bsAb1 and its aggregation, Protein A (PA) micro purification was carried out on the Tecan Fluent 1080 using PhyNexus tips packed with Affinity MabSelect SuRe matrix (Cytiva). The PhyNexus tips, with a resin bed volume of 40 µL, had a binding capacity of 390 µg. Four milliliters of supernatants were transferred to two 96-well plates, and an affinity chromatography was performed on the Tecan Fluent. The Buffer A (20 mM Na-Citrate, 20 mM Na-Phosphate, pH 7.5) was used for the equilibration and washing, and Buffer B (20 mM Na-Citrate, pH 3.0) for elution. The eluate was collected into a 0.5 mL deep-well plate and neutralized with 0.5 M Na₂HPO₄, pH 8.0, with a dilution factor of 1:10. The eluate was analyzed by analytical size exclusion chromatography (aSEC) and capillary electrophoresis (CE-SDS).

Chromatographic instruments and methods

Combination of capture and separation steps on the ProteinMaker

The purification process using the ProteinMaker instrument (Protein BioSolutions) is a sequence of PA capture and CEX separation steps.

For the PA capture step, the column used was the 5 mL HiTrap® MabSelect™ SuRe™ (Cytiva). Buffer A (20 mM Na-Citrate, 20 mM Na-Phosphate, pH 7.5) was used for equilibration and washing of the column, and Buffer B (20 mM Na-Citrate, pH 3.0) for step elution.

For the CEX separation step, a 1 mL POROS™ GoPure™ XS Pre-packed was used (Thermo Fisher Scientific). Buffer C (20 mM Na-Phosphate, pH 7.0) was used for equilibration, washing of the CEX column, and was mixed with Buffer D (20 mM Na-Phosphate, 400 mM NaCl, pH 7.0) for the gradient elution. A total of 22.5 mg of protein, derived from supernatants from CHO transient expression, were loaded onto the 5 mL MabSelect™ SuRe™ column at 1 CV/min. After a 5 CV wash with Buffer A, proteins were eluted with 4 column volumes (CV) of Buffer B. About 8.5 mL corresponding to the protein elution was collected in a 24-well plate and immediately loaded onto the POROS™ GoPure™ XS 1 mL column for CEX at 1 CV/min. The gradient of the CEX chromatography consists of 16 elution steps with increasing concentration of sodium chloride as indicated in Table 1. Fractions of 2.3 mL were collected into four 96-well plates. The concentration measurements were carried out with the Lunatic reader (Unchained labs) integrated into the Tecan system. The hitpicking (100 µL per fraction), the normalization to a maximum concentration of 0.5 mg/ml, and the sample preparation for the aSEC and CE-SDS analysis were performed using the Tecan Fluent 1080. All worklists were generated by the Roche Autolab platform.

Table 1.

Elution profile of the CEX chromatography method performed on the ProteinMaker.

Fraction	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
Elution buffer (%)	0	4	8	12	16	20	24	28	32	36	40	44	48	52	56	100
[NaCl] (mM)	0	16	32	48	64	80	96	112	128	144	160	176	192	208	224	400

Preparative size exclusion chromatography with the ÄKTA pure/autosampler

The polishing pSEC step was performed using an ÄKTA Pure (Cytiva) system equipped with an ALIAS autosampler (Cytiva) and a 16 × 600 mm Superdex 200 pSEC column (Cytiva). The CEX pool was transferred into a U-shaped glass vial for injection via the ALIAS autosampler into the column. The pSEC column was pre-equilibrated with the Buffer E (20 mM Histidine, 140 mM sodium chloride, pH 6.0) at 1 mL/min for 140 mL. The protein sample was applied to the column at a flow rate of 1 mL/min, and the eluate was collected in 1 mL fractions into a 96-well plate. The selected fractions based on aSEC testing were pooled and 0.22 µm sterile filtered. After filtration, 120 µL from the final container were collected to complete quality control of the product.

Cleaning-in-place procedure

Before each use, the PA chromatography column and the CEX column were sanitized with 10 CV of 0.1 M sodium hydroxide at a flow of 0.5 CV/min. Every week, the pSEC column was flushed with 1 CV of 0.1 M sodium hydroxide at a flow rate of 1 mL/min and kept in this state for at least 12 h. After each sample injection, the sample pathway including the autosampler loop and the fraction collector tubing was flushed with 0.1 M sodium hydroxide for 50 mL at a flow rate of 5 mL/min.

Analytical methods for protein characterization

Titer determination

The protein concentration in the cell culture supernatant was determined using an analytical PA titer determination method on an Agilent Infinity II HPLC system equipped with a UV detector. The column utilized was an Upchurch 2 × 20 mm column with an internal volume of 63 µL, freshly packed for each run with POROS 20 A/G resin (Applied Biosystems). Equilibration and washing steps were performed with Buffer A (10 mM Tris, 50 mM Glycine, 100 mM NaCl, pH 8.0), while elution was achieved using Buffer B (10 mM Tris, 50 mM Glycine, 100 mM NaCl, pH 2.5). The elution was performed as a step elution with 100% Buffer B. The process operated at a flow rate of 1 mL/min, with a total run time of 3.5 min. The UV detector monitored signals at wavelengths of 280 nm, 260 nm, and 215 nm. Peaks were integrated (area under the curve) using the Agilent OpenLab CDS software.

Analytical size exclusion chromatography

The Product peak SEC, High molecular weight (HMW) and Low molecular weight (LMW) contents were determined using a TSKgel UP-SW3000 column on an Agilent Infinity II HPLC system with a UV detector. The running buffer was 0.025 M K₂HPO₄, 0.125 M NaCl, 0.2 M L-Arginine, 0.02% NaN₃, pH 6.7. The method was executed at a flow rate of 0.3 mL/min, with a total duration of 10 min. A 25 µL injection of normalized samples, at a maximum concentration of 0.5 mg/mL for in-process control or 12 µg for final quality control, was loaded onto the column. Signals were recorded at wavelengths of 215 nm and 280 nm. Peaks were integrated (area under the curve) using the Agilent OpenLab CDS software.

Capillary electrophoresis

The purity of the samples was evaluated through CE-SDS analysis under both reducing and non-reducing conditions. CE-SDS analysis was conducted using the HT Protein Express Reagent Kit (Revvity) on the LabChip GX Touch system (Revvity). Samples (2 µL of normalized samples at a maximum concentration of 0.5 mg/mL for in-process control or 2 µL at 0.5 mg/mL for final quality control) were mixed with 7 µL of Protein Express Sample Buffer, with or without reducing solution, heated at 70°C for 10 min, and then diluted with 35 µL MilliQ water. Seven hundred and fifty microliters of HT Protein Express Wash Buffer, provided with the reagent kit, was prepared, and the chip was set up according to the manufacturer’s instructions. The HT Antibody Analysis 200 method was used to analyze the samples, and the results were processed with LabChip GX Review software to determine molecular weight and purity.

A280 concentration determination

A volume of 2.5 µL of samples and buffer blanks was transferred from a 96-well plate to a Lunatic plate for UV absorbance measurement at 280 nm using a DropSense 16 instrument (Unchained Labs). The protein concentrations were determined using the molar extinction coefficient based on the amino acid sequence.

Endotoxin analysis

Samples were initially diluted in endotoxin-free water at a fivefold dilution factor for storage prior to measurement. Subsequently, they were further diluted in Acila glass tubes using LAL Reagent Water at a 1:10 dilution ratio for analysis. Endotoxin analysis was conducted using the Limulus Amebocyte Lysate (LAL) test, performed with the Kinetic-QCL Bulk Kit (Lonza). The analysis utilized the WIN-KQCL System (Lonza), comprising a BIOTEK microplate reader. LAL Reagent Water (Lonza) was used throughout the procedure. All materials were certified endotoxin-free, including microplates, pipette tips, and glass tubes.

Results

Platform overview

The platform was designed for the preparation and delivery of high-quality, asymmetric IgG-like molecules referring in this study to bispecific antibodies (bsAbs) and antibody fusion proteins. The objective was to combine the purification process, in-process control strategy and automated liquid handling, all overseen by a digital solution orchestrating the workflow and storing data in the Laboratory information management system (LIMS) and data warehouse (Figure 1).

The first stage corresponds to the quality assessment of the expressed proteins. First, an analytical PA chromatography analysis provides titer information for the protein of interest. Next, a PA micropurification is performed, and the collected eluate is characterized by aSEC and CE-SDS to assess the correct pairing of the molecule and the aggregation content. Based on the results, an appropriate purification strategy is selected. It can be a two-step (PA capture and polishing pSEC) or a three-step process with an additional separation step (CEX) between the capture and the polishing.

The second stage is the purification process performed on our semi-automated platform. Proteins of interest are grouped according to the purification strategy, to enable simultaneous purification on the ProteinMaker. The Autolab workflow controls the process, from ProteinMaker purification to analysis and pooling of fractions, in preparation for the polishing step. The purification process ends with the polishing pSEC.

The third stage is the final quality control of the purified material, followed by aliquotation and centralized storage to facilitate accessibility for our researchers. Finally, the operations are documented in the Electronic lab notebook (ELN).

Starting material side products profile defines the purification strategy

The semi-automated process is suitable for a variety of asymmetric antibody formats, presented here is the evaluation of a 1 + 1 bsAb named bsAb1. The format is characterized by potential side products with similar molecular weight, making it a perfect example to assess the efficiency and robustness of our platform (Figure 2a). The first stage of the process involves the quality assessment of the expressed bsAb1, which will subsequently be purified. The 1 + 1 IgG-like bsAb1 is captured by a PA micro purification, and the eluate is analyzed by aSEC and CE-SDS (Figure 2b).

The aSEC chromatogram profile (Figure 2b) provided a product peak quantification of 61.7% for the protein of interest. In addition, the profile showed 18.4% of HMW impurities and 19.9% of LMW impurities. According to the target molecule format and its potential side products (Figure 2a), it is not possible to discriminate the target molecule from side products with a similar molecular weight. The monomer content is therefore potentially lower than this estimation. This was confirmed by in-process liquid chromatography-mass spectrometry (LC-MS) analysis highlighting the presence of significant amounts of the side product A2HK (light chain mispairing) and traces of the side product ABK2.

The composition of the eluate is characterized by capillary electrophoresis performed in non-reducing conditions (Figure 2b), indicating the purity of the product peak (75%). Several side products (25%) with a lower molecular weight than the target molecule were also observed. They indicated the presence of products lacking light chains and of half molecules. As for the aSEC, it is not possible to discriminate the target molecule from any side products with a similar molecular weight.

Under reducing conditions, the heavy and light chains were observed as expected (Figure 2b). It is not possible to assign the peaks to the respective chains. Therefore, the electropherogram has been annotated with HC1 and HC2 for the heavy chains and LC1 and LC2 for the light chains. While on this electropherogram, the poor separation between LC1 and LC2 does not allow precise estimation of the chain ratio under reducing conditions, their relative comparison is an important information to assess the purity of the correctly paired molecule during the separation step of the purification process (Table 2). The ratios between the peak area LC1/LC2 (2.97) and HC1/HC2 (1.23) indicate the presence of an excess of certain chains, such as LC2, while the intended chain stoichiometry for the target molecule is 1:1:1:1.

Table 2.

Batch information and biochemical characterization of the eluates (fractions 1 to 11) from the CEX chromatography. The fractions 6 to 8 highlighted in the frame were selected for pooling. The selection of the fraction was based on the product peak (%) in aSEC, the purity of the product peak (%) in non-reduced (NR) CE-SDS and the ratio of the peak area HC2/HC1 and LC2/LC1 in reduced (R) CE-SDS. All data except the chain ratios in CE-SDS R (in gray) are captured and stored in the LIMS or data warehouse via Autolab.

Fraction	1	2	3	4	5	6	7	8	9	10	11
Elution Buffer (%)	0	4	8	12	16	20	24	28	32	36	40
[NaCl](mM)	0	16	32	48	64	80	96	112	128	144	160
Batch ID	−13	−14	−15	−16	−17	−18	−19	−21	−22	−23	−24
Concentration (mg/mL)	0.44	1.03	0.29	0.23	0.17	0.66	1.25	0.71	0.18	0.13	0.08
Volume (mL)	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3
Amount (mg)	1	2.4	0.7	0.5	0.4	1.5	2.9	1.6	0.4	0.3	0.2
aSEC
HMW %	33.1	14.9	13.1	25.6	34.5	13.8	5.6	15	61.5	51.3	22.9
Product peak SEC %	5.8	41	64.2	51.6	46.8	83.2	91.3	75.4	28	20.4	17.6
LMW %	61.1	44.1	22.7	22.8	18.7	3	3.1	9.6	10.5	28.3	59.5
CE-SDS NR
Purity product peak %	4.8	44.8	75.1	73.5	71.8	92.8	96.4	90.1	89.3	74.8	44.3
CE-SDS R
Ratio HC2/HC1	0.3	2.7	1.7	1.8	1.4	1.1	1	1.2	1.2	1.3	1.7
Ratio LC2/LC1	44.3	25.4	18.7	9.6	4.8	1.2	1	1.2	1.3	1	1

In addition to the quality assessment, the titer was measured by analytical PA chromatography at 95 mg/L.

The characterization of the starting material provides valuable insights into the amount and type of side products and supports the selection of appropriate purification methods. For the molecule bsAb1, the data generated and the potential presence of side products with similar molecular weight justify the addition of a separation step and result in a three-step purification process. The PA capture step and a CEX chromatography are performed sequentially on the ProteinMaker platform and orchestrated via Autolab. The third step, namely the polishing pSEC, is performed on an Äkta Pure connected to an Alias autosampler.

Purification of the 1 + 1 IgG-like bsAb on the semi-automated platform

Description of the three-step purification process

The ProteinMaker instrument is equipped with 12 PA and 12 CEX chromatography columns. The setup enables the combination of two sequential purification steps (PA capture and CEX separation) with the direct injection of the PA eluate (pH3.0) into the CEX chromatography column. A single run can handle 12 harvested samples in five to six h depending on the sample volume, typically 0.2 to 0.4 L.

After analysis of the CEX elution fractions, a CEX elution pool is generated and subsequently purified using pSEC. Twelve samples can be injected consecutively through the pSEC column using the Alias Autosampler. This three-step purification process, PA – CEX – pSEC, enables the efficient purification of up to 12 asymmetric antibodies in less than 3 d.

Autolab is a key part of the laboratory operation system and controls the process from the harvest container to the registration of the CEX pool

An Autolab workflow orchestrates the process from the configuration of the ProteinMaker instrument until the pooling of the CEX fractions. It provides a unified interface where a Business Process Model and Notation (BPMN) (Figure 3) coordinates the tasks and communicates with the different parties involved in the fulfillment and registration of the operations such as the equipment (ProteinMaker, Liquid handler, Analytical devices), the digital systems (LIMS, data warehouse) and the operator.

Figure 3.

The figure displays three parts of the Autolab workflow orchestrating the purification process with their corresponding Business process modeling and notation (BPMN). The first part (a) corresponds to the capture of the UV signal from the CEX elution, its transformation and display as a histogram plot. A minimum UV threshold can be set in order to pre-select “hits” (highlighted in blue). The operator can edit the selection manually or modify the threshold to refine the selection. This part ends with the registration of the elution plates with the “hits” as new purification batches in the LIMS. The second part (b) involves the liquid handling system. It starts with the sampling of the hits into a hitpicking plate. This is followed by concentration measurement, normalization of the samples, and preparation of the in-process control (IPC) analytics (aSEC and CE-SDS). After performing IPC analytics and reviewing the data, the operator can create the pooling selection in Autolab (c) and automatically generate a worklist for the liquid handling system. The figure displays the pooling plate registered in the LIMS and the container information available ensuring traceability of the sample.

The device’s integration (Liquid handler, ProteinMaker, Lunatic, HPLC and CE-SDS instruments) enables system controls and automated data acquisition. The operator can review data that have been automatically transformed to facilitate decision-making. For example, the UV trace generated by the ProteinMaker is transformed into a histogram plot (Figure 3a) displaying the relative protein concentration of each eluted fraction from the CEX chromatography and enabling the selection of “Hits” that will be registered and further characterized by concentration measurement and biochemical analysis (aSEC and CE-SDS) (Figure 3b). Based on the operator’s decision for “Hits” or “Pooling” selection, Autolab automatically generates worklists for the liquid handling system enabling plate preparation.

Through the Autolab orchestration, the platform can exclusively handle samples formerly registered in the LIMS with a valid identifier (ID) and ensures that newly registered samples are defined by unique identifiers and barcodes. Each intermediate batch and container is registered with quality attributes within a barcoded plate, increasing the granularity of the data generated and ensuring the traceability of the sample along the process. Finally, Autolab transfers the experiment documentation in pdf format to the ELN.

The three-step process allows removal of side-products to obtain > 98% final product purity

The ProteinMaker purification results in the generation of 16 CEX elution fractions. Based on the histogram plot provided by Autolab, fractions 1 to 11 were selected as “hits” and registered as purification intermediate batches for further characterization (Figure 3a). The results are displayed in Table 2.

In order to select the fractions containing the 1 + 1 IgG-like bsAb with the highest purity, we focused on four experimental data. The purity of the product peak measured by capillary electrophoresis and the product peak SEC measured by aSEC ensured the efficient purification of the molecule with the desired molecular weight. The ratios HC2/HC1 and LC2/LC1 determined by capillary electrophoresis in reducing conditions confirmed the correct pairing of the purified molecule (Table 2).

The fractions 6 to 8 had the highest values for the product peak percentage in CE-SDS and aSEC. The profile of the remaining side products in aSEC for these fractions was compatible with their removal during the polishing step. For these fractions, the ratios HC2/HC1 and LC2/LC1 were both in a range [1–1.2] confirming the proper pairing of the molecule. The total protein amount of 6 mg, obtained after pooling of the fractions (Figure 3c), represented a recovery of 26.6% from the initial load of supernatant. The quality attributes measured for the CEX pool are shown in Table 3.

Table 3.

Purification overview for the molecule bsAb1. All data except the recovery (highlighted in gray) are captured and stored in the LIMS or data warehouse.

Quality assessment	Expression	CEX pool	Final pool
Batch ID	−12	−45	−46
Concentration (mg/mL)	0.09	0.91	1.19
Volume (mL)	250	6.6	2.3
Amount (mg)	22.5	6	2.7
Recovery (%)		26.6	12
Yield (mg/L)			10.8
CE-SDS NR
Purity (%)	74.7	96.3	99.1
aSEC
Product peak (%)	61.7	88.4	98.3
HMW (%)	18.4	7	1.2
LMW (%)	19.9	4.2	0.5
LC-MS			Identity confirmed
Endotoxin (EU/mg			<0.21

The polishing step was pSEC, which allowed the separation of the remaining side products and exchanged the target molecule to an appropriate formulation buffer to ensure optimal storage conditions. The quality attribute threshold was defined as greater than 95% product peak in aSEC and greater than 90% purity in non-reduced CE-SDS based on the impact of the different side products on the future experiments. HMW and LMW impurities should be limited to a maximum of 3% in aSEC in the final purified product due to their potential impact on the biological activity in an in vitro or in vivo assay. The quality attributes of the CEX pool were already close to these specifications.

In order to mitigate the risk of exceeding 3% of HMW and/or LMW impurities in the final product, the pooling strategy for the pSEC was to select all fractions presenting a product peak percentage above 97% in aSEC. After the polishing step, the final purification batch was concentrated and registered. The quality attributes measured for the final pool are shown in Table 3. The mass of the purified molecule was confirmed by LC-MS and no side products presenting a similar molecular weight as the target molecule bsAb1 were detected. The final pool represents a production yield of 10.8 mg of purified product per liter of supernatant and a recovery of 12%. The ProteinMaker purification was repeated three times using different purification channels. The CEX pool (fractions 6–8) showed an average amount of 5.54 mg (standard deviation 0.59 mg) and a product peak SEC average percentage of 87.8% (standard deviation 1.46%).

Platform evaluation: the semi-automated purification platform leads to a three-fold productivity increase while maintaining high-standard quality

The purification platform was not intended to purify standard antibodies but was exclusively dedicated to the purification of various bispecific IgG-like antibody formats and antibody fusion proteins. To assess the performance of our semi-automated process, over 500 purified molecules were monitored using several quality attributes (product peak SEC, CE-SDS purity, endotoxin concentration) and performance indicators (production yield, productivity). The molecules were produced by transient expression in a standardized volume of 0.2 to 0.4 L.

After the platform implementation, the final quality attributes of 89% of the purified proteins showed an aSEC product peak above 95% (median Mdn 98.3%, Inter Quartile Range IQR (96.8–99.3)). For 84% of the proteins, the product peak measured by CE-SDS was above 90% (Mdn 98.5%, IQR (94.9–100)). The endotoxin concentration was measured below 1EU/mg for 92% of the released batches, indicating suitability for in vivo experiments (Mdn 0.18 EU/mg, IQR (0.11–0.33)). Finally, the production yields were over 2 mg/L for 92% of the bsAbs purified with an average value of 17.6 mg/L (Mdn 11.8 mg/L, IQR (5.5–22.6)). These data were compared with those of 400 purifications performed before the platform implementation. A Mann–Whitney statistical test was performed to compare the data, taking into consideration that the data are not normally distributed. The results showed no statistical difference between the two groups for the product peak in aSEC (p-value 0.3162). For the three other categories, the results of the Mann–Whitney test showed statistical differences (p-value <0.0001) with higher purification yield and purity measured by CE-SDS since the implementation of the new process. The endotoxin values increased slightly but are mostly below the threshold of 1 EU/mg. The results indicate that the increase in throughput was achieved while maintaining the high quality of the proteins produced (Figure 4).

Figure 4.

Evaluation of the semi-automated purification platform for the purification of bispecific antibodies and antibody fusion proteins. The figures display and compare the outcome of 510 purifications after the implementation of the semi-automated purification platform to 407 purifications prior implementation. For each data set, the graphics show the median value and the interquartile range (IQR). The increase in throughput was achieved while maintaining the high quality of the proteins produced and with improved yields.

The use of the semi-automated purification platform and its workflow orchestration software Autolab also increased the amount of asymmetric antibody purifications per operator per week from 4 to 12, resulting in a three-fold increase in productivity. For a variety of projects and formats (bispecific, trispecific, antibody fusion proteins), the platform delivered sufficient material for our portfolio project requirements with minimal process adaptation (e.g., change to a pH gradient for the CEX elution or use of hydrophobic interaction chromatography (HIC) method).

Discussion

The platform enhances the supply of asymmetric antibodies for pre-clinical drug discovery

The ProteinMaker was shown to be fast and efficient for the purification of monoclonal antibodies as described in the study by Yang et al.¹⁴ However, the purification of asymmetric antibody formats, referring in this study to bsAbs and antibody fusion proteins, remained hindered by the need for an additional chromatography step to separate mispaired or incomplete molecules. The successful transfer of our CEX method on the ProteinMaker resulted in a three-fold increase in productivity while maintaining our quality standards. The platform described in this study enabled the production of asymmetric therapeutic antibodies with an average yield of 17.6 mg/L of supernatant, facilitating multiple assays and in-depth characterization.

The implementation of separation steps, CEX or alternatively HIC, on the system was challenging due to the technical limitations of the equipment. With only one pump per purification channel, the system did not allow linear gradient elutions. This step was therefore replaced by a succession of 16-step elutions. The choice of 16 fractions was a compromise in order to optimally cover our gradient range and fit our platform for a variety of antibody formats, but also to limit the number of fractions generated in the context of multiple parallel purifications. This resulted in the generation of 192 fractions for the purification of 12 molecules in parallel.

The UV detection signal from the ProteinMaker provided information on the protein amounts and permitted a first selection of the fractions. The number of samples to be further characterized was limited to 95 per run and, in consequence, restrained the overall cost and duration of the process. The automation of liquid handling tasks and the systematic characterization of the samples enabled the generation of standardized and complete data sets. The unified interface provided by Autolab streamlined data management by automating the retrieval of input information, the capture and registration of outputs, and the documentation of the process. This ensured traceability and allowed operators to concentrate on making informed decisions.

The ability to fulfill tasks overnight (e.g., aSEC, pSEC queues) enabled, on one platform, the three-step purification of 12 complex asymmetric antibody formats in 3 d. This allowed the scheduling of three staggered purification processes in a week, representing a maximum throughput of 36 molecules with three operators. The increase in productivity leads to new opportunities, such as shorter supply cycles for drug development and the possibility to evaluate more binder combinations and formats.

Further improvements will focus on increasing the throughput and the production yield

The purification process described above is semi-automated. The operator is still required to execute manual tasks and make decisions. For example, transferring the various plates between the ProteinMaker, the liquid handling system, and the analytical devices is not automated, preventing immediate transitions between stages. Additionally, the analytical data must be carefully reviewed to select the correct fractions, as different formats will have different side product profiles. Some limitations persist due to the wide variety of constructs purified on the platform. The process requires adaptations for certain constructs with specific physicochemical properties (e.g., low isoelectric point) or for side products that exhibit similar charge profiles to the target molecules. To address these issues, alternative methods such as CEX with pH gradient elution or HIC have been developed to ensure the production of these molecules with the desired quality. If these adjustments cannot be anticipated, additional chromatography steps may be required. This extra work can increase the operator’s workload, negatively affect yield, and slow down the process.

Finally, the sample handling after the pSEC must not be overlooked. These tasks involve potential concentration, sterile filtration, quality control sample preparation, and aliquoting for final storage. The workload increases significantly when managing multiple samples simultaneously. Therefore, operators would greatly benefit from automation capabilities for these tasks as well.

Improvements in the upstream process are not directly related to the semi-automated system presented here but could enhance its performance by increasing the production yields. Low production yields can negatively impact project costs and timelines if reproductions are needed to obtain sufficient material. Three factors play an important role: 1) the titer of the supernatant, 2) the correct assembly of the target molecule during the expression and its aggregation content, and 3) the number of purification steps needed to obtain the desired quality attributes. A potential improvement could therefore be an increased titer, which can be achieved by optimizing expression plasmid design, transfection conditions, or by using different hosts (CHO or HEK). Additionally, optimizing the quality of the expressed protein is crucial. A high amount of mispaired or incomplete proteins can negatively impact separation efficiency and potentially reduce the capacity of the columns. An optimized plasmid ratio could enhance both the quality of the expressed protein and the titer and is usually determined for repeated or scaled up supplies.

The complexity of the design space for the transient production of asymmetric antibodies makes its outcome unpredictable and necessitates human intervention to take the appropriate decisions and ensure process control. The digital integration of the semi-automated purification platform described in this work provides strong data governance and is the basis to build large- and high-quality data sets. The data captured are various and complex and can take various forms, such as text (e.g., cell host, purification process, buffers), numbers (e.g., concentration, product peak aSEC and HMW content) and images (e.g., chromatogram, electropherogram). The combination of these data and their relative importance constitutes the logic behind decision-making throughout the process. The progress observed in computational technologies creates opportunities for the analysis and valuation of large data sets and can lead to the generation of accurate models.²⁶ These new insights will result in better process understanding and predictability, laying the foundations for increased productivity.

Funding Statement

The author(s) reported that there is no funding associated with the work featured in this article.

Disclosure statement

The authors are Roche employees and disclose ownerships of Roche holding shares.

Abbreviations

a280

absorbance at 280 nm

aSEC

analytic size exclusion chromatography

bsAb

bispecific antibody

CE-SDS

capillary electrophoresis SDS

CEX

cation exchange chromatography

CV

column volume

ELN

electronic lab notebook

HC

heavy chain

HIC

hydrophobic interaction chromatography

HMW

high molecular weight

IgG

immunoglobulin G

IPC

in-process control

IQR

inter quartile range

LC

light chain

LC-MS

liquid chromatography mass spectrometry

LIMS

laboratory information management system

LMW

low molecular weight

Mdn

median

NR

non reducing conditions

PA

protein A

pSEC

preparative size exclusion chromatography

QC

quality control

R

reducing conditions