Antibody Engineering And Therapeutics

Getting From a Hybridoma to Something That Actually Works In Vivo

The whole Antibody Engineering And Therapeutics pipeline is less glamorous than the papers make it look. You sequence a gene, you clone it, you express it, and then you spend the next six months watching your protein aggregate in buffers that looked perfect on paper. I have seen people spend more time troubleshooting formulation than they did designing the construct. Most people jump straight into variable region cloning without thinking about the human framework. You need to pick a scaffold that actually tolerates the CDR grafting you plan to do. Framework residues outside the CDRs aren't just structural decoration. They make direct contacts with antigen in many cases, and if you transplant a murine CDR onto the wrong human framework, you will lose most of your binding affinity through subtle changes in the electrostatic surface potential. I worked on a program where we grafted a high-affinity rat CDR3 into a standard IgG4 human framework. The SELEX hit had nanomolar affinity. The engineered humanized version came back at 50 micromolar. We spent three months trying to fix it before someone finally suggested reverting framework residue 71 back to the rat glutamine. That single change recovered 80 percent of the binding. Nobody predicts that from the structure. You just have to do it and measure it.

Affinity Maturation Is Not Just Directed Evolution

Error-prone PCR gets a lot of use in academic labs. It works fine when you are starting from a microgram of plasmid and have nowhere else to be. But in a therapeutic context you need controlled mutagenesis. Site-saturation mutagenesis at the CDR positions followed by yeast display screening gives you far better return on investment. You can cover all twelve CDR positions with reasonable depth in a single experiment, whereas random mutagenesis buries you in the thousands of non-functional or neutral clones. Here is something most people do not factor in. The highest affinity clone is not necessarily the best lead. A clone at 1 nanomolar often has worse biophysical properties than one at 10 nanomolars. Higher affinity usually means slower off-rates, and slow off-rates correlate with higher viscosity at formulation concentrations. I have seen candidates get dropped from development because the 1 nM binder required 150 mg/mL formulation and the injectable limit was 100 mg/mL. The 8 nM version was ten times easier to formulate and had identical pharmacodynamics in the mouse model.

Expression Systems: Choosing Between Them Matters More Than You Think

HEK293 cells give you better glycosylation patterns for therapeutic antibodies. CHO cells are cheaper and scale easier. The choice is not just about cost. If you are engineering for reduced Fc effector function, you need to worry about fucose addition. CHO cells add fucose by default. HEK293 does not, or does so at very low levels. If your construct depends on afucosylation for enhanced ADCC, using CHO cells means you have to engineer the cell line to knock out the FUCT enzyme. That adds months to your timeline. For bispecific antibodies, the challenge multiplies. T-cell engagers in particular suffer from chain mispairing. Heavy chains pair with heavy chains from the wrong arm, light chains pair with light chains from the wrong arm, and you end up with a soup of four products instead of the desired heterodimer. The knobs-into-holes mutation scheme helps but does not solve the problem completely. You typically get 60 to 70 percent correct pairing after Protein A purification. The rest requires an additional cation exchange step or size exclusion, which eats yield. I ran into this with a BiTE construct where the two heavy chains had nearly identical isoelectric points. The ion exchange column just would not separate them. We ended up switching to a streptavidin-based capture step that exploited a biotin tag on one of the chains. It was an ugly workaround that cost extra reagents, but it gave us 95 percent purity in a single step. We never did figure out why the charge difference was so small. The sequence looked different enough that it should have been separable.

Glycoengineering for Constant Region Function

The Fc glycan at Asn297 is a master regulator of antibody function. Core fucose reduces CD16A binding affinity by roughly tenfold. This is not a small effect. For therapeutic antibodies that rely on ADCC, like many oncology candidates, removing fucose is mandatory. The commonly used Chimeric Saccharomyces cerevisiae platform produces human-like glycans with zero fucose. It is well established but the culture conditions require tight control. Glucose feeding rates shift the glycan profile toward high-mannose species if you are not careful, and high-mannose glycans trigger mannose receptor clearance in the liver. There is also the question of sialylation. Adding terminal sialic acid residues can reduce Fc effector functions and may be useful for inflammatory indications where you want the antibody to bind antigen without recruiting immune cells. But sialylated antibodies have shorter half-lives in some cases due to clearance through the asialoglycoprotein receptor if the sialylation is incomplete. You need fully sialylated product, which means tighter process control.

Stability Testing That Actually Predicts Behavior

Many labs measure thermal stability by DSC and report Tm values. A Tm of 70 degrees Celsius sounds good until you realize your antibody is aggregating rapidly at 40 degrees Celsius in formulation buffer. Thermal midpoint and colloidal stability are only weakly correlated. You need to run kinetic stability studies at 37 and 40 degrees Celsius in your actual formulation buffer, not in PBS. PBS is not a realistic formulation environment for any therapeutic antibody. Polar aromatic residues in the heavy chain constant region, particularly tyrosine at position 337, are hotspots for aggregation. These residues are surface exposed and prone to forming transient hydrophobic clusters that nucleate aggregation. I have seen mutations at this position improve stability by twentyfold without any change in antigen binding. The same residue is present in most human IgG1 frameworks, which is why so many IgG1 therapeutics struggle with long-term storage stability.

Immunogenicity Assessment Before You Invest In Preclinical Studies

In silico T cell epitope prediction tools like NetMHCpan are reasonably accurate but they have blind spots. They miss conformational epitopes and they perform poorly on MHC class II predictions, which matter more for antibody immunogenicity. The standard approach is to deimmunize by mutating predicted epitopes, but every mutation you introduce carries a risk of affecting binding or stability. I learned this the hard way with a candidate where we removed what we thought were three redundant epitopes. Two of those mutations shifted the hydrogen bond network in CDR2 and dropped affinity by a factor of five. The practical workaround is to make the minimum number of changes necessary to disrupt the epitope anchor residues, and to keep as much of the native sequence as possible. You do not need to remove every predicted epitope. You need to remove the ones that are actually presented efficiently by common HLA alleles in the target population. If your drug is for a niche indication with a small patient population, the HLA binding profile matters differently than for a mass-market indication.

Manufacturing Considerations That Get Deferred Until It Is Too Late

Downstream processing is where a lot of promising antibodies die. If your antibody precipitates at pH 5 during the low pH viral inactivation step, you cannot simply adjust the buffer. The precipitation is often irreversible. I have seen teams spend weeks trying to find a formulation that keeps the antibody soluble at pH 4.5 and then realize the excipient they added interacts with the Protein A resin and reduces capture yield. It is a connected system and every parameter affects every other parameter. Aggregates are the primary concern. Even subvisible particles below 10 micrometers can trigger immune responses. The industry standard for reporting is to count particles above 10 micrometers and above 25 micrometers, but the real risk is in the 1 to 10 micrometer range where standard light obscuration methods have poor sensitivity. You need to use flow imaging or particle analysis by suspension cytometry if you want accurate data on the damaging size range. Many groups skip this because it adds cost, and then they get surprised during regulatory review. The whole field of Antibody Engineering And Therapeutics moves faster now than it did ten years ago. The tools are better. The libraries are bigger. But the fundamental problems remain the same: your sequence does not always fold the way you expect, your construct does not always behave the way the data predicts, and the step that fails is almost always the one you least expected to fail.