In vitro immune safety for biologics: cytokine release, immunogenicity and ADA assays

In vitro immune safety assessment uses human cell-based assays to evaluate whether a candidate biologic could trigger an unwanted immune response, such as excessive cytokine release or the development of anti-drug antibodies (ADAs).

For a monoclonal antibody, bispecific, immune cell engager, antibody-drug conjugate (ADC) or Fc-fusion molecule, this work sits alongside the pharmacology package and feeds the safety case in an investigational new drug (IND) application or clinical trial application (CTA).

It carries particular weight for immune-modulating biologics, because the animal species used in toxicology assessments can differ from humans in ways that matter for immune activation, and because antibody responses observed in animals are generally poor predictors of immunogenicity in humans.

This guide covers which assays are relevant and when, how a cytokine release assay (CRA) is designed, which in vitro systems inform immunogenicity risk, what an ADA assay measures, what animal studies can and cannot tell you and how to plan the work with an outsourced partner CRO.

Which immune safety assays does a biologic need, and when?

There is no standard immune safety testing package for biologics. ICH S6(R1) states that routine tiered testing approaches and standard testing batteries are not recommended for biotechnology-derived pharmaceuticals, and ICH S8, which covers immunotoxicity studies for human pharmaceuticals, excludes biotechnology-derived products from its scope. The assays are instead selected from the mechanism of action (MoA) and the risks it implies.

The FDA's guidance on the nonclinical evaluation of the immunotoxic potential of pharmaceuticals (June 2023) takes the same position: understanding the MoA is important in determining which assays are appropriate, and the assays suited to a T cell-engaging molecule differ from those suited to a checkpoint inhibitor or a receptor agonist.

The cellular distribution of the target and the molecule's potential to activate or modulate key immune cell populations, such as T cells and dendritic cells, should guide the selection.

In practice, three questions shape the package for an immune-modulating biologic.

• Can it activate immune cells inappropriately? This is the cytokine release question, and it is most pressing for molecules that cluster or crosslink receptors on immune cells.
• Will patients develop antibodies against it, and will those antibodies matter clinically? This is the immunogenicity question. Risk can be assessed using a range of in vitro approaches before dosing, while ADA assays are used clinically to determine whether an antibody response has developed.
• Is the chosen toxicology species informative? Comparative work in human and animal-species systems addresses this and is covered in our guide to the in vitro data package from lead selection to IND.

Timing follows the same logic. Screening-level work belongs in lead optimisation, where it can still change the candidate. Fit-for-purpose studies supporting the first-in-human dose and the monitoring plan belong in the IND-enabling phase.

When is a cytokine release assay needed?

A cytokine release assay is generally considered where the MoA creates a plausible route to immune cell activation: T cell engagers and other immune cell engagers, agonist antibodies against co-stimulatory receptors, molecules with enhanced Fc effector function and antibodies against targets expressed on immune cells.

The clinical concern behind it is cytokine release syndrome (CRS), the systemic inflammatory response that can follow rapid, large-scale cytokine production after dosing. Assessing that risk in vitro, before first-in-human dosing, is what the assay is for.

The FDA's 2023 immunotoxicity guidance notes that where a product has already demonstrated a clear potential to cause cytokine release, such as a CD3 bispecific T cell redirector, additional assays may not be necessary because the hazard is already identified. Similarly, where one assay format is positive, running a second format may not add anything.

It also sets out what a positive result means. A positive CRA does not necessarily preclude further development; depending on the indication, the magnitude and duration of the effect and the cytokines involved, it may instead shape the starting dose, the clinical monitoring plan and the dose escalation and stopping criteria.

The reason this assay exists in its current form is the 2006 first-in-human trial of TGN1412, a humanised superagonist anti-CD28 antibody. Suntharalingam et al. reported that all six volunteers who received the antibody developed a systemic inflammatory response within 90 minutes, despite a nonclinical package that had not predicted it.

Two key findings helped explain the gap, and both still inform assay design.

Stebbings et al. demonstrated that the response could be reproduced in vitro when the antibody was immobilised rather than presented in soluble form, a difference the same group went on to characterise in detail.

Eastwood et al. demonstrated that CD28 is not expressed on the CD4+ effector memory T cells of the macaque species used in preclinical testing, so those cells could not respond to a CD28 superagonist at all.

How do you design a cytokine release assay?

The main design choices are the cell source, how the biologic is presented, the donor panel, the controls and the cytokines measured. The FDA's 2023 guidance describes the standard formats as soluble and immobilised presentation using peripheral blood mononuclear cells (PBMCs), whole blood and other matrices, and lists endothelial cell co-cultures, tumour and endothelial co-cultures and patient-derived cells as further options.

The assay format can have a major impact on the response that is observed. Vessillier et al. compared platforms directly and reported that a PBMC solid-phase assay detected the TGN1412 response with a very small donor group, while whole blood assays needed group sizes an order of magnitude larger to reach the same statistical power. In contrast, alemtuzumab-induced cytokine release was detected much more sensitively in whole blood.

The practical conclusion is that format should follow mechanism, and that a negative result in an unsuitable format does not clear the risk.

• Cell source. Whole blood keeps red cells, granulocytes and plasma proteins in the system. PBMCs remove them, which can be what allows a response to be seen. Fresh and cryopreserved material can produce different response profiles, making consistency of sourcing and handling an important consideration during assay development.
• Presentation. Soluble, immobilised on the plate or presented on a cell. For molecules that act by clustering receptors, presentation can be the difference between detecting a response and missing it entirely.
• Pre-culture. Römer et al. demonstrated that a short pre-culture of PBMCs at high cell density can restore responsiveness to soluble TGN1412, with CD4 effector memory T cells identified as the main source of the proinflammatory cytokines. High-density pre-culture can help cells recover from isolation-induced stress and restore responsiveness in certain assay formats.
• Donors. Responses vary between individuals, so a single donor establishes very little. A tiered approach screens formats in a smaller donor set, then confirms in the most relevant format across a wider panel if necessary.
• Controls. The FDA guidance states that assays should include appropriate positive and negative controls, with clear measurable outcomes, and that information should support the assay being fit for purpose. Vessillier et al. reported an international collaborative study establishing the first reference antibody panel for this purpose, comprising positive antibodies based on published anti-CD52, anti-CD3 and superagonist anti-CD28 sequences with matched isotype controls.
• Cytokines. Commonly measured cytokines include interleukin-2 (IL-2), IL-6, interferon-gamma (IFN-γ) and tumour necrosis factor-alpha (TNFα), although the optimal panel depends on the mechanism of action and the biological questions being addressed. The FDA's 2014 immunogenicity guidance recommends as broad a panel as practical.

Surveys of practice by Finco et al. and the workshop reported by Grimaldi et al. both found that multiple platforms remain in use, that the choice depends on the MoA, throughput and the controls and donors available and that more data on the predictive performance of each platform is still needed.

For CD3 bispecifics specifically, the workshop summarised by Kamperschroer et al. identified excessive cytokine release, on-target off-tumour activity and neurotoxicity as the key clinical safety concerns.

At RoukenBio we typically use a tiered approach to cytokine release assays. Initial screening is performed across multiple assay formats in a smaller donor panel, before moving into confirmatory studies in the format most relevant to the molecule's mechanism of action, if necessary. Depending on the target biology, these studies may be performed using whole blood, PBMCs or both.

Which in vitro assays assess immunogenicity risk?

The assays in common use are dendritic cell maturation, T cell activation and proliferation, MHC-associated peptide proteomics and in silico epitope prediction. They are not used to predict clinical ADA incidence; what they can do is rank candidates and formulations against each other and identify the sequence and product features that drive a T cell response.

The EMA's guideline on immunogenicity assessment of therapeutic proteins states that non-clinical studies aiming to predict immunogenicity in humans are not normally required.

It does encourage ongoing consideration of emerging in silico, in vitro and in vivo approaches as development tools or for a first estimate of risk, and notes that assays based on innate and adaptive immune cells could help reveal cell-mediated responses.

Most of these assays are designed to assess pathways associated with T cell dependent immunogenicity. As Jawa et al. set out in their updated consensus review, T cell epitopes within the sequence of a protein therapeutic drive or modulate the ADA response, depending on the type of T cell response they stimulate.

Each assay addresses a different point in that pathway.

• Dendritic cell maturation. Monocyte-derived dendritic cells are exposed to the biologic and assessed for surface marker expression, such as CD83, CD80, CD86 and CD40, cytokine secretion and signalling. This reports innate activation and antigen-presenting cell engagement.
• T cell activation and proliferation. PBMC-based or dendritic cell to T cell co-culture formats measure CD4+ T cell responses, using activation markers such as CD154, CD134 and CD137, cytokine secretion and proliferation readouts. Joubert et al. reported that responses in an assay of this type, run across up to 50 naive donors, showed a relationship with the observed clinical immunogenicity of the ten antibodies tested.
• MHC-associated peptide proteomics (MAPPs). This identifies which peptides from the biologic are actually presented on human leukocyte antigen (HLA) class II, and is, as Karle notes, among the best characterised of these assays for immunogenicity assessment.
• In silico epitope prediction. Sequence-based analysis is a fast first pass that narrows what the cell-based assays need to test.

Two design points decide whether the data is usable. The first is donor coverage: because the response depends on HLA type, a representative panel is needed.

Cohorts of approximately 30–50 donors are commonly used for in vitro T cell immunogenicity assays, depending on the molecule, assay and risk assessment. That contrasts with functional primary immune-cell assays, which can begin with three or more donors to establish activity and donor-to-donor variability.

The second is benchmarking. Results are interpretable mainly in relation to controls and to biologics with known clinical behaviour.

Ducret et al., reporting for the European Immunogenicity Platform's non-clinical working group, describe keyhole limpet haemocyanin (KLH) as a widely accepted positive control because of its high neo-antigenic content, and recommend including a low-immunogenicity and a medium-to-high benchmark alongside it.

Their review also found that protocol variability between laboratories makes the risk assigned to a given candidate hard to compare, which is an argument for holding your own assay conditions constant across a programme.

The value of benchmarking becomes clear when considering the wide range of immunogenicity observed across marketed biologics. Ridker et al. reported that 48% of patients who received bococizumab had detectable anti-drug antibodies at one year, and that neutralising antibodies developed in 29% of all patients who received it. The mean reduction in low-density lipoprotein cholesterol was progressively smaller in the patients with the highest titres, falling to 12.3% in the top decile.

At the other end of the range, the US prescribing information for bevacizumab reports treatment-emergent anti-bevacizumab antibodies in 0.6% of patients in adjuvant solid tumour studies.

Our immunogenicity and safety assessment services cover dendritic cell maturation, T cell functional assays and dendritic cell assays, with the donor strategy, benchmarks and decision criteria agreed before the first experiment.

What is an anti-drug antibody assay, and when is it needed?

An anti-drug antibody assay detects and characterises antibodies that bind to a biologic in patient or animal samples. It answers a different question from the in vitro work above: not whether a biologic is likely to be immunogenic, but whether antibodies are present in a given sample, at what titre and whether they neutralise the drug.

Unlike immunogenicity risk assessment assays, ADA assays are typically developed as product-specific methods intended to support regulatory decision making throughout clinical development. Their purpose is not to predict whether antibodies might occur, but to detect and characterise them when they do.

Why it matters is set out in the FDA's immunogenicity assessment guidance (August 2014). Neutralising antibodies can block efficacy by binding domains critical to it, and both neutralising and non-neutralising antibodies may alter pharmacokinetics by increasing clearance or, conversely, by prolonging half-life.

The guidance frames the whole assessment as risk-based and case-by-case.

The assays run in tiers. The FDA's ADA assay guidance (January 2019) describes a sensitive screening assay applied to all samples, a confirmatory assay to establish that screen-positive samples contain antibodies specific to the product, and then titration and, where relevant, neutralisation assays on the confirmed positives. The EMA guideline describes the same multi-tiered structure.

Three parameters from that guidance tend to drive assay development.

• Cut points. A low but defined false-positive rate of approximately 5% is described as desirable for the screening assay, because it maximises detection of true positives; a 1% false-positive rate is described as adequate for the confirmatory cut point.
• Sensitivity. The guidance recommends that screening and confirmatory IgG and IgM assays achieve a sensitivity of at least 100 ng/mL, noting that greater than 100 ng/mL may be acceptable depending on risk and prior knowledge, and that neutralisation assays may not reach that level.
• Drug tolerance. Circulating drug can mask antibodies, so the guidance recommends examining drug tolerance early in development, and notes that sampling at trough and techniques such as acid dissociation can help.

On timing, the FDA guidance recommends that the immunogenicity risk assessment and the rationale for the testing paradigm are provided in the original IND. It encourages testing in phase 1 and phase 2 with suitable screening and confirmatory assays, and states that samples from pivotal studies should be tested with fully validated assays.

Where immunogenicity poses a high clinical risk, assays may be needed before clinical studies begin, with testing in real time.

Pre-existing antibodies are worth checking before the clinic. As Gorovits et al. describe, these are immunoglobulins that bind or cross-react with the biologic in treatment-naive subjects, and their clinical consequences vary from none to significant, which makes them a risk factor to evaluate rather than a finding to interpret in isolation.

RoukenBio runs ADA ELISA workflows in screening, confirmatory and neutralising formats, supported by our bioanalytical team. Our research poster on detecting pre-existing anti-drug antibodies in treatment-naive human serum sets out a surface plasmon resonance (SPR) based approach to that question.

What do animal studies tell you about immune safety?

For predicting human immunogenicity, animal studies are generally less informative than is sometimes assumed. ICH S6(R1) states that the induction of antibody formation in animals is not predictive of a potential for antibody formation in humans, and its 2011 addendum states that such analyses in nonclinical studies are not relevant in terms of predicting the immunogenicity of human or humanised proteins in humans.

What ADA measurement in animals can do is support interpretation of the animal study itself. ICH S6(R1) recommends measuring antibodies in repeat-dose toxicity studies to aid interpretation, characterising the response and correlating it with pharmacological and toxicological changes, and the addendum identifies altered pharmacodynamic activity, unexpected changes in exposure and evidence of immune-mediated reactions as the triggers for doing so.

For cytokine release, the FDA's 2014 guidance is explicit that data from animal studies and in vitro assessments may supplement each other but are generally not fully predictive of the clinical outcome.

It also states that in vitro assessments of cellular activation in human whole blood or PBMCs help overcome known limitations of animal models for some T cell subsets, and that signs of cellular activation in vitro should be considered a potential clinical risk even when animal findings are negative.

This is also the direction of current FDA policy. The April 2025 roadmap to reducing animal testing in preclinical safety studies begins with monoclonal antibodies, notes that animal immunogenicity is not predictive of human immunogenicity and cites the development of in vitro cytokine release assays after TGN1412 as a case where human cell-based methods added a safety net.

In September 2026 the FDA issued a direct final rule clarifying that non-animal methods can be used where appropriate, while stating that it does not eliminate or prohibit animal studies or change evidentiary standards.

The roadmap groups this work under New Approach Methodologies (NAMs), which it describes as encompassing in vitro human-based systems, in silico modelling and other platforms, including ex vivo human tissues and high-throughput cell-based screening. Cytokine release, immunogenicity risk and ADA assays in human cells sit inside that definition.

Animal studies remain important where they are scientifically justified and provide information that cannot be readily obtained elsewhere. However, human-relevant in vitro and ex vivo approaches are increasingly being used alongside animal studies to strengthen confidence in clinical translation.

Taken together, these guidance documents point to a consistent conclusion: animal studies remain valuable for understanding pharmacology and toxicology in a relevant species, but they are generally poor predictors of human immunogenicity and may not fully predict immune activation risks.

Where does target antigen density come in?

For a targeted therapy, the amount of antigen expressed on a cell is often a key determinant of whether a drug can act on a cell, which makes antigen density part of the immune safety question as well as the efficacy question. A candidate biologic that triggers cytokine release at the low target densities found on healthy tissue carries a different risk profile from one that needs tumour-level densities.

This is assessed directly by running the functional assay against target cells set to a series of defined densities, including the densities measured on relevant healthy cells. Running the same assay against target-negative cells is how in vitro work addresses off-target immune activation.

Our on-target, off-tumour risk profiling uses the IndEx-2 inducible cell line platform for this. A published example is the study of CLN-978 by Meetze et al., which used inducible CD19-expressing lines to establish the CD19 density at which the engager was active.

The method is set out in RoukenCells' guide to target antigen density and the therapeutic window, and in RoukenCells' guide to inducible expression cell lines. Understanding the antigen density required to trigger activity helps define the therapeutic window and assess the likelihood of on-target activity in healthy tissues.

How do you choose a partner for immune safety studies?

Look for a CRO that works routinely in primary human immune cells, that will help design the study rather than only run it and that reports data in a format that can be used directly in regulatory documentation.

Questions worth asking:

• Which cytokine release formats do you run, and how would you choose for my candidate? The answer should start from the MoA, not from the format the laboratory prefers.
• How many donors do you use as standard, and how do you handle donor variability? Expect different answers for screening and for confirmatory work.
• What positive and negative controls and benchmark molecules do you include? Reference materials and clinically characterised benchmarks make results comparable.
• What are the acceptance criteria, and are they set before the study runs? Criteria agreed afterwards are hard to rely on.
• Can you cover immunogenicity risk and ADA bioanalysis as well? Keeping the immune safety package with one team reduces the gaps between pieces of it.

Direct experience with your modality matters more than organisational scale alone. The relevant experience for an immune cell engager is not the same as for an Fc-silenced antibody or an ADC.

No single assay can determine whether a biologic will be immunogenic in the clinic. Instead, these approaches provide complementary information that helps build a weight-of-evidence assessment of risk.

RoukenBio works this way as a preclinical immunology CRO. Our immunotoxicology and in vitro pharmacology teams design the study with you, run it in primary human immune cells and patient tissue, and report it in a format that can go straight into a package. Our preclinical services page sets out the full scope.

The most informative immune safety studies are those designed around the biology of the molecule rather than a predefined testing package.