Question 1: Which bispecific formats have you made?

Format experience is the first filter, because each format family brings a different impurity profile and process route. A CDMO that has only made symmetric IgG-scFv constructs has never removed heavy-chain homodimers; one that has only made Fc-containing formats has never designed a non-Protein A capture.

Ask for experience by family rather than molecule count: asymmetric IgG with CH3 heterodimerization, symmetric appended IgG, fragment-based constructs without an Fc, and two-cell-line processes based on controlled Fab-arm exchange. Ask about T-cell engagers specifically: low clinical doses mean small batches, high-sensitivity analytics and a co-culture potency assay. Request anonymized summaries: format, host, scale reached, clinical use.

The answer you want is a described problem and its solution, not a list of logos.

Question 2: What is your cell line platform?

The right answer covers host choice, chain-ratio control and assembly-based clone screening, not just a titer figure. The clone that secretes the most protein is the wrong clone if a third of it is homodimer, and bispecific antibody CDMO services should show how they avoid selecting it.

• Host options. CHO-K1 and CHO-S are the common defaults; CHO-DG44 supports DHFR-based selection; HEK293 suits transient material; microbial hosts fit small fragment formats.
• Chain-ratio control. Ask whether chains sit on one vector or several, and how each chain's expression is tuned.
• Assembly-based screening. Ask which method scores heterodimer content at the clone stage: non-reduced CE-SDS, HIC or intact mass, supported by a dual-antigen ELISA.
• Clonality and stability. Ask for clonality evidence (imaging at single-cell deposition) and the stability study length, as ICH Q5D expects.
• Cell banking. Ask whether cell banks are made under cGMP on site, how storage is monitored and whether off-site backup is offered.

Question 3: How do you remove mispaired species?

Mispair removal is where bispecific antibody CDMO services earn their margin, so ask for the toolbox and for proof that it worked at scale; even well-engineered asymmetric formats secrete homodimer and half-antibody.

A credible toolbox includes pH-gradient Protein A elution when one arm carries Protein A-ablating mutations, cation exchange exploiting the pI differences of charge-pair designs, hydrophobic interaction and mixed-mode resins, and kappa- and lambda-specific affinity resins for κλ designs. Detection matters equally: nobody can optimize a polishing step for a species they cannot quantify, so ask which methods were run on development fractions.

Finally, ask how the step behaved on scale-up. Gradients that resolve homodimer on a 1 mL column can collapse on a 60 cm column unless packing, residence time and buffer accuracy are controlled, and the yield penalty of tighter pooling belongs in the batch economics.

Question 4: What is in your analytical package?

A bispecific analytical package should include every monoclonal antibody method plus heterodimer-specific identity, purity and two-arm potency methods, validated by the group that will run release testing.

Ask who develops, validates and transfers the methods to QC; a split across three groups or vendors is the most common source of release delays in bispecific antibody CDMO services.

Question 5: What is the scale path to 2,000 L?

Ask for the equipment ladder from bench to commercial scale and the logic behind it, because a bispecific's lower yield makes the path longer: half the yield of a standard IgG means twice the bioreactor volume or twice the batches.

Three points separate strong bispecific antibody CDMO services from weak ones. First, continuity: the same organization should run the 3–5 L development bioreactors, the 50 L pilot runs and the GMP trains, so scale-down models reflect real equipment. Second, the stainless-steel versus single-use choice should be explained by batch size, changeover and cost, not offered as a default. Third, downstream capacity must keep pace: column diameters, UF area and buffer volumes sized for a bispecific's extra chromatography cycles.

Question 6: What is your inspection record?

Inspection history with the agencies you will file with is the best predictor of a smooth pre-license inspection. Ask what each certification means and when it was last renewed.

• PIC/S GMP. Certification by an authority in the Pharmaceutical Inspection Co-operation Scheme means inspection against GMP standards harmonized with EU GMP; it supports, but does not replace, inspection or MRA-based reliance by the agencies you file with.
• US FDA. Ask whether the site was inspected for an approved application and how observations were closed.
• Japan PMDA and EMA. Ask about accreditation, inspection history and qualified-person audits if Japanese or EU supply is planned.

Ask for the dates of the last inspections and the number and class of observations. Bispecific antibody CDMO services without a dated record are a risk, however strong the science.

Question 7: How are IP and technology transfer handled?

Clarify who owns the cell line, the process and the data before the first vial is thawed; bispecific programs carry more third-party technology than monoclonal antibodies.

The test of good bispecific antibody CDMO services is whether the exit package lets another site reproduce the process without the original scientists.

Question 8: What drives timeline and cost?

Bispecific programs cost more and take longer than monoclonal antibody programs for identifiable reasons, so ask bispecific antibody CDMO services to itemize them. The drivers are the number of cell lines (two for Fab-arm exchange processes), lower titer, extra polishing steps with lower yields, a larger analytical package with two potency assays, and longer characterization and stability programs. Each should appear as a line in the proposal with its assumptions.