Lonza [SWX: LONN]

Peter O’Callaghan

Introduction

Many of today’s expression platforms for recombinant protein manufacturing have been optimised for the expression of monoclonal antibodies (mAbs), often with the platform fundamentals originating from the 1990’s. However, as new modalities increase in prevalence, and wider changes in the biopharmaceutical sector such as the use of automation and AI-based drug design accelerate this trend, there is a need to design more flexible production paradigms. A case in point is the growing prevalence of multi-chain asymmetric antibodies and bispecifics (bsAb) in development, which demand cross-platform improvements from expression vector design through to purification and analytical process optimisation to achieve economically-viable product yields. As pressures on manufacturing platforms increase, there is a need not only to develop new expression tools, but also to re-purpose and re-optimise existing ones. In this short article we’ll take a brief journey through some of the efforts that Lonza has made to optimise the manufacturability of bsAbs and other formats, and take a glance towards future directions with an emphasis on the upstream aspects.

Upfront Design, from DNA and Protein Sequence

Manufacturability considerations ideally need to be baked into the product as early as possible in the discovery process, with appropriate input into molecular design such that expression challenges can be mitigated early on. That being said, manufacturability considerations and application of associated developability tools should not constrain bsAb molecular design upfront, but rather be used to enable the output from discovery groups by providing solutions that are ‘plug-and-play’ in nature and have broad applicability. Examples of such solutions include the knobs-into-holes technology developed by Genentech, which solves the HC-HC heterodimerisation challenge through amino acid substitutions in the heavy chain CH3 domains, and Lonza’s bYlok® technology, which relocates the existing VH-VL disulphide bridge to the CH-CL interface to prevent HC-LC shuffling in the Fab arms of asymmetric bsAbs.

Even with the most ‘exotic’ of molecular designs, expression platforms such as CHO still retain the necessary biosynthetic ‘chops’ to generate economically[1]viable production titres. However, much like any human needs an instruction manual or training course to learn a new piece of lab equipment, the set of genetic instructions supplied to the cell at transfection needs to be continually updated to better enable the expression of molecules that production hosts have not evolved to synthesise at multi[1]gram industrial scales. 

Mammalian expression vectors have remained broadly unchanged in overall structure for decades, although recent developments in the field of linear DNA, such as the use of linear double-stranded DNA with covalently closed ends, suggest that significant disruption could potentially be on its way. For now, we can focus on improved expression control of the different bsAb chains, achieved via the use of engineered promoters and associated elements such as UTRs, or more subtle system tweaks such as the use of alternative gene orders and/or additional gene copies. At Lonza we have found that the latter approaches can often yield significant benefits not only on titre but also on product quality, although in the case of gene order effects, the relative lack of predictability means that empirical testing is required on a case-by-case basis.

The complex interplay between expression vector design, clone expression titre, long-term expression stability, and product quality, is an area where artificial intelligence (AI) will surely make a game-changing impact. However, to leverage these new in silico tools it will be necessary to build out large datasets based on high[1]throughput screening of expression vector variants for a wide-range of different molecular formats, and to obtain relevant multi-level phenotypic data on the resultant clones. Expression vector design spaces may be reduced upfront by the use of DoE (design of experiments), albeit at the risk of missing out on potentially important data points. That being said, the winners in this expression vector design ‘race’ will be those who best leverage customer networks as well as data from internal commercial services work to build AI models based on high quality data from industry-relevant production formats.

Converting Genetic Instructions Into Product Titres

So, we’ve given the CHO host cell line an optimised set of genetic instructions for product synthesis, but as we know from any human team given a work task, interpretations of the brief can vary. In this context, differences in expression vector integration site within the CHO genome, coupled with stochastic process effects will likely still require extensive clone screening. Within Lonza we have streamlined the CLC process over the years, particularly for bsAb expression, via the on-boarding of new high[1]throughput (HTP) technologies such as the Beacon™ optofluidic system for clone isolation and screening, supported by follow-up scale-down fed-batch miniature bioreactor assessment to check for platform fit of the resultant clones.

"As pressures on manufacturing platforms increase, there is a need not only to develop new expression tools but also to re-purpose and re-optimize existing ones"

Despite these improvements, the current emphasis is still very much on leveraging HTP screening and empirical testing of clones. Again, this is an area where AI-based predictive tools can potentially be disruptive, for example by the development of models that accurately predict the platform performance of clones while at the single cell stage or shortly thereafter as initial colonies. Single-cell multi-omics methods are rapidly gaining in maturity, and logically it should be possible to predict which clones will generate the highest, stable titres at such early stages without long-term empirical testing, using models built upon extensive omics data to find the ‘omics fingerprint predictive of performance at industrial scale.

Clearly the rate of change in the industry is set to significantly accelerate. The challenge will be to keep doing what we do best as an industry, while capturing the value from legacy and future datasets to ensure that the revolution initiated by ChatGPT is fully utilised to bring about the step-change in industrial performance that we need to meet the growing demands of patients