There’s no one way to make bispecific antibodies — it depends on which platform you use! The process is consistent, but the places where platforms differ are worth understanding, as they might influence your choice in one direction or another.
This quick guide explains that the biggest differences between platforms appear in the assembly step, where issues like light-chain mispairing can create downstream problems with yield, purity, scalability, and formulation. We then show how our B-Body platform avoids those issues structurally, enabling cleaner screening, more reliable characterization, and a more manufacturable bispecific.
If you want a deeper comparison of the platform families themselves, our guide to bispecific antibody platforms covers that ground. If this is the first you’re hearing of the B-Body platform for bispecifics, get the 101 here.
Here’s a quick rundown of how bispecific discovery works in general.
Before anyone finds an antibody, they need the target in a form that can be screened. That means cell lines and purified protein that present the target the way it appears in the human body.
If the reagent doesn't represent the real thing, everything downstream inherits the error. Many teams looking for a bispecific have done part of this already while evaluating whether the target idea was viable.
Two broad routes exist in this stage:
In vivo methods immunize an animal (usually a humanized mouse or a rabbit) and harvest what its immune system produces. Rabbits are popular because they generate antibodies unusually quickly, which significantly reduces the immunization wait time.
In vitro display methods skip animals entirely and screen enormous synthetic libraries in a dish. Phage display is the workhorse. Yeast display is often used alongside it.
Neither route is inherently better than the other. Display gives you more control over your starting material and avoids inheriting the quirks of an animal's immune response. Immunization can produce binders that are hard to find synthetically.
Most of the industry does something in this space, and the steps look broadly similar across discovery teams.
Here is where the paths diverge, and this stage is why bispecifics are harder to produce than monoclonals.
An antibody is composed of four chains—two heavy and two light. In a normal monoclonal, both halves are identical, so there is only one correct way for them to come together.
A bispecific antibody has two different arms, so four different chains have to find the right partners. Two things can go wrong. The two heavy chains can pair with themselves instead of each other, and the light chains can attach to the wrong arm. The heavy chain problem is largely solved. Knobs-into-holes, a bump engineered on one heavy chain and a matching groove on the other, has been used across the industry for years and works well. (We use it too!)
The light chain problem is where platforms differ, and the approaches split into a few families: skip the problem with a shared common light chain, steer the chains with engineered charges, or change the architecture so the arms can't pair incorrectly. Fragment-based formats sidestep the question entirely by not building a full antibody.
You must separate the molecule you want from everything else the cells made, then find out whether it works. How hard that is depends on what you did in stage three.
Every IgG-like bispecific must address the same two fundamental challenges: pairing the correct heavy chains and preventing light-chain mispairing.
Bispecific formats like knobs-into-holes and common light chain work, but they can create manufacturing or compatibility compromises. Poor chain pairing leads to lower yields, higher impurities, and longer development timelines.
Combinatorial diversity of bispecific IgGs. Overview of possible combinations to arrange heavy and light chains from two different antibodies, including strategies to overcome incorrect heavy-light chain pairing.
Here’s the same process, run end-to-end, on Invenra’s B-Body platform.
Invenra’s Rapid Discovery services provide an efficient, industry-leading approach to discovering monoclonal and bispecific antibodies. With cutting-edge phage display technology and our streamlined workflow, we deliver high-quality antibody leads in record time.
We evaluate available reagents and pick the ones that best present the target, so the libraries have something realistic to bind to. Cell lines, purified protein, whatever the target requires.
This is where Invenra expertise shines. Our primary workflow begins with phage display, using proprietary libraries holding so many possible sequences that you need scientific notation to describe the contents of a single tube.
Mechanically, it’s less mysterious than it sounds. The target is attached to a magnet, the candidate antibodies are added, and you see which ones stick. The craft is in the rest of it: washing away everything that binds non-specifically and getting as much sequence diversity as possible from a single shot.
The part that is genuinely different is the parallelization.
We separate our libraries based on physical characteristics, such as antibody size and parental sequence, and keep them separate rather than pooling everything.
Then we query them against multiple screening conditions at once, because nobody knows in advance which condition best represents what happens in a patient. That gives multiple shots on goal, both across libraries and across conditions.
The alternative, as it’s done elsewhere, is one tube at a time. Run it, see what you get, and if it isn't what you needed, go back to the beginning and repeat. Projects run that way take six to twelve months.
Material coming out of phage, ideally still millions of unique sequences, often moves into yeast display. Yeast gives more dynamic range and more dimensions along which to sort individual antibodies, so you can ask sharper questions than phage alone allows.
Everything from every selection and screen then goes into next-generation sequencing. The output tells us which sequences came out of which condition in which well, which turns a lot of data into a map.
From that sequencing data, we choose around 400 sequences and express them as actual antibodies from mammalian culture.
Then the characterization phase begins, and this is where we would argue the difference shows up most. It’s not just "does it bind?”
It covers whether:
It binds anything it shouldn't (polyreactivity)
Its surface properties are amenable to manufacturing (hydrophobicity)
It’s assembling the way it should
It binds the analogous target in your safety toxicology species
The panel has epitope diversity, which we establish through binning experiments to see whether two antibodies stick to the same place or different ones
We spend time on this because we see issues when people bring us work done elsewhere. Frequently, what they received was a panel of antibodies and one experiment showing binding to the target, and that was it. That’s “tip-of-the-iceberg” characterization, and it can make results look better than they are, because they have not looked beneath the waterline.
Now you have characterized monoclonals against target A and characterized monoclonals against target B, and you want bispecifics.
You pick the best monoclonal antibodies for each target, and the B-Body platform lets us build every possible permutation of those combinations rather than guessing at a few. A small-scale expression and purification pipeline isolates each permutation cleanly, in its final bispecific format.
Example functional data as a heat map representing the screen
Example SEC data in a 12x12x2 matrix
Then you go through the monoclonal characterization phase a second time, except now with bispecifics. Same assays, same readouts, same rigor.
That repetition is the whole point. We’re using tried-and-true methods that we can only run because of what the B-Body platform gives us. We don't have to worry about things like light chain mispairing. We can run assays that tell us how much impurity is present, because nothing is hiding underneath.
A platform that produces mispaired species muddies its own analytics. If a meaningful fraction of what you made is the wrong molecule and you can't cleanly resolve it, your purity number is essentially a guess. Removing that problem is what lets well-understood methods work on a multispecific.
Function trumps everything. We start with the functional data, select the top performers, and then cross-reference them against developability assays. Of the top hundred, perhaps thirty carry no serious developability red flags, and those move forward.
This is a consultative review, not a simple data handover. You see the data, and we work through it together.
Nobody goes from a one-milliliter high-throughput experiment to a conversation with regulators. Selected candidates are scaled up to 200 mL, supporting multiple purification steps and yielding many milligrams of protein.
At that scale, you can dig properly into the assays and validate what the screen suggested.
Check out slides 15-20 of our talk at AET for a map of the two entry points—our own mAb discovery and partner-supplied antibodies via B-Body Express—converging into bispecific discovery.
This deck also breaks out the discovery steps individually.
Everything above describes discovery. The decisions, whether they are good or bad, follow the molecule into manufacturing.
This is the failure mode that causes the most trouble. If light chains mix and match, three things happen.
First, you may not have a competent molecule, because the wrong light chain paired with a given heavy chain may not bind properly.
Second, batched are not consistent, so you have to prove potency within guardrails, typically around +/-5%. If pieces of your molecule scramble, you risk failing batches of drug.
Third, and worst: a molecule where the light chains are flip-flopped is almost impossible to purify, because it is essentially identical to your drug. There is no clean separation method for something that similar.
On the B-Body and T-Body platforms, we don't see light chain shuffling, and we achieve that through the architecture rather than by adding charge engineering on top, which is also why the platform works with whatever antibody pair you choose.
There’s a second element worth mentioning. The knob-and-hole region can produce a small amount of knob-knob or hole-hole homodimer. On our platform, that isn't a problem, because those species differ enough in charge and other properties to be separated.
Yield is mostly a function of how complicated your purification is. Every chromatography step costs you molecule volume, and in early phases you can expect to lose around 20% per step even in a good case.
So the cost of a shuffling problem isn't only the shuffled material. It's the extra chromatography steps you need to clean it up.
Add one and you've cut another 20%.
Add two, and you may be down 40%. That comes directly out of what you can put in a bottle and ship to your Phase 1 trial.
With B-Body, we scale it up and process our bispecific molecules like monoclonal antibodies. B‑Body bispecifics antibodies typically deliver 6 to 11 g/L (the green bar in the graph below) from stable CHO cell lines. For comparison, a panel of other bispecific platforms averaged 2.4 g/L (grey). This is the highest‑expressing bispecific platform we're aware of, and it drops right into standard mAb‑like processes.
B-Body bispecific yield (green) against a panel of other bispecific platforms (grey). B-Body runs 6–11 g/L; the panel averages 2.4 g/L.
Purity and yield are the same conversation from different directions, because you can’t administer until you hit specification, generally 95% purity or better. If you can reach that in two chromatography steps, great! If you need a third, you have to run it, because until you hit spec, your batch has failed.
Those steps are not cheap. An additional step is not a scheduling inconvenience. It can be a seven-figure line item in the cost of goods.
B-Body reaches dosing-grade purity in two chromatography steps. We achieve >80% purity in a single purification step and >95% in a standard two‑column process. The figure below demonstrates purity >95% for all formats, including 1×1, 2×2, 2×1, one‑arm, and an IgG1 control.
That single-step purity holds across formats: after polishing, size-exclusion purity lands at 90 to 95% or better across 1×1, 2×1, and 2×2.
Size-exclusion purity by format, before polishing (grey) and after (green), across 1×1, 2×2, 2×1, trispecific, one-arm, and an IgG1 control. After polishing, purity lands at 90–95% or better.
What kills scalability is having to do anything unusual. If you have to tell a CDMO they need a chromatography method they've never run, for example, that alone can kill a project.
Being able to treat B-Body molecules as if they were monoclonal antibodies and still achieve the purity and yield means they run as what the industry calls a platform process. That's roughly code for "we don't have to work very hard to give you what you want," and it's a compliment. Most facilities can turn a platform process to thousands of liters without breaking a sweat.
Concentration is the constraint here. To dose subcutaneously, a molecule must survive high concentrations without precipitating, and some formulations struggle badly.
Camelid-derived single-chain platforms are useful for comparison because they tend to be highly hydrophobic. Past a certain concentration, those molecules start sticking to each other, and once that begins, you have a snow globe in a tube and the product is finished.
There’s real engineering effort going into fixing this. We’ve seen presentations showing re-engineering camelid antibodies are less prone to aggregation, with reports reaching 10 mg/mL. When we make B-Body antibodies, ten to twenty times that concentration is within range.
We attribute this to the use of fully human antibody sequences and to the B-Body architecture itself. The payoff for patients is the difference between a subcutaneous injection and sitting in a chair for an hour or two on an IV.
One structural detail deserves explaining because it does most of the heavy lifting.
On a B-Body antibody, only one arm carries a CH1 domain. That makes CH1 a purification handle: an anti-CH1 resin grabs only chains carrying it, and by expressing the other chains in excess, you stack the deck toward pulling out the correctly assembled molecule.
Here’s a look at the B-Body:
The B-Body scaffold. (1) knobs-into-holes Fc so the heavy chains pair with each other; (2) proprietary CH3 domains replacing CH1/CL in one Fab arm, so the right light chain pairs on its own; (3) plug-and-play variable domains from any source; (4) a sole CH1 domain that allows one-step anti-CH1 purification.
Where the anti-CH1 handle matters most is at the beginning, during screening. You can’t develop a bespoke purification method for every one of hundreds or thousands of candidate multispecifics.
It would take far too long. Anti-CH1 gets every one of them to "clean enough to run an assay" quickly and uniformly. That’s what makes matrix screening practical.
It can also be used at scale-up, added at the front to improve initial product quality and make later steps easier. We flag that as the more debatable use, because the industry norm is a standard starting path that doesn't include anti-CH1, and deviating from the norm is exactly the kind of unusual thing that complicates a CDMO relationship. It's an option, not a default.
Bispecific antibodies are made by finding good binders, combining them into one molecule, and then proving the result is manufacturable and does what you need. The industry broadly agrees on the first and last parts.
What differs is the middle, and the consequences of that difference arrive late, in purification steps you didn't budget for, batches that fail specification, and molecules you can't concentrate enough to dose the way you wanted.
Our approach with the B-Body platform is to eliminate the assembly problem structurally, so the rest of the process can be straightforward. Tried-and-true methods, run twice, on molecules that behave.
Talk to us
Have a target pair in mind? Tell us what you're trying to build and where you are, and we'll walk you through what the process would look like for your program.
Contact us »
Or start with the data: Our B-Body explainer covers the yield, purity, formulation, and matrix screening data behind the process described here. Learn more about the B-Body Platform »