Adalimumab (Humira®) is a fully human monoclonal antibody that targets tumor necrosis factor-alpha (TNF-α), a pro-inflammatory cytokine central to the pathogenesis of various autoimmune and chronic inflammatory diseases (1, 2). Since its initial approval, Humira has become one of the most widely prescribed biologics worldwide, with approved indications that include rheumatoid arthritis, Crohn’s disease, ulcerative colitis, psoriasis, and ankylosing spondylitis (3). Following the expiration of market exclusivity, development of adalimumab biosimilars has accelerated to expand patient access and reduce treatment costs (4–8). As of July 2025, the U.S. Food and Drug Administration (FDA) has approved ten adalimumab biosimilars: Amjevita® (adalimumab-atto), Cyltezo® (adalimumab-adbm), Hyrimoz® (adalimumab-adaz), Hadlima® (adalimumab-bwwd), Abrilada® (adalimumab-afzb), Hulio® (adalimumab-fkjp), Yusimry® (adalimumab-aqvh), Idacio® (adalimumab-aacf), Yuflyma® (adalimumab-aaty), and Simlandi® (adalimumab-ryvk), with additional candidates in development.

Regulatory approval of biosimilars requires a rigorous demonstration of analytical comparability to the reference product across physicochemical, structural (9), and functional attributes (10–14). Among the functional assays for anti-TNF-α monoclonal antibodies, cell-based TNF-α neutralization assays are widely used for lot release and stability testing. A collaborative effort to establish the first international bioactivity standard for adalimumab highlighted the diversity of binding and functional assays employed across laboratories, including our group at the FDA (15). Binding has been evaluated by direct ELISAs using immobilized TNF-α with detection via HRP-conjugated anti-IgG (Fc-specific), anti-IgG1, or anti-kappa chain antibodies; alternative platforms included electrochemiluminescence (ECL), fluorescence resonance energy transfer (FRET), bio-layer interferometry (BLI), surface plasmon resonance (SPR), and flow cytometry using CHO cells expressing a membrane-bound, non-cleavable form of TNF-α (16). TNF-α neutralization was commonly measured as inhibition of TNF-α-induced cytotoxicity in murine fibroblast (L929) (10) or fibrosarcoma (WEHI-164 and WEHI-13) (17) cell lines. Other approaches included reporter gene assays (RGAs) using HEK-293 cells transfected with NF-κB-responsive reporters (luciferase or SEAP) and apoptosis assays quantifying TNF-α driven caspase activation in U937 cells, wherein neutralizing antibodies reduced the apoptotic signal (18).

While no single assay addresses all characterization needs, RGAs are particularly valuable for evaluating the biological activity and consistency of anti-TNF-α antibodies, including adalimumab (19). By directly measuring NF-κB activation in response to TNF-α and its inhibition by therapeutic antibodies, RGAs provide a sensitive, and MoA-reflective readout suited to lot release, stability monitoring, and biosimilar comparability studies.

In this study, we qualified a NF-κB reporter gene assay to measure the neutralizing activity of adalimumab and its biosimilars. The assay was evaluated for working range, specificity, precision (repeatability and intermediate precision), reproducibility, and system suitability, in concordance with USP chapters < 1032>(20) and < 1034>(21), as well as ICH Q2(R2) (22) on bioassay design, data analysis, and method validation/qualification. We further demonstrated the assay’s ability to quantify activity and support similarity assessments between reference adalimumab and its biosimilars.

Cell seeding. Assay-ready A549-IkB cells were thawed and seeded into sterile, tissue culture treated 96-well plates at 9,600 − 10,000 cells/well and incubated at 37°C with 5% CO₂ for 48 hours prior to assay.

Antibody and TNF-α preparation. On the assay day, serial dilutions of adalimumab were prepared in a separate 96-well dilution plate (two replicate rows per article). A fixed concentration of TNF-α (3.0 ng/mL) was added to each adalimumab dilution and pre-incubated for 30 minutes at room temperature. In parallel, a TNF-α activity curve was prepared by serially diluting TNF-α from 200 to 1.9x10^− 4 ng/mL using a 1:3 scheme.

Assay Execution and Detection. Using a multi-channel pipette, 20 µL from each well of the dilution plate was transferred to the assay plate containing A549-IkB cells. Plates were incubated for 2-hours at room temperature. Detection reagents containing the complementary β-galactosidase fragment were added and incubated for 15 min, followed by addition of the β-galactosidase substrate and a 1-hour incubation to generate luminescence from the reconstituted active enzyme. Luminescence (RLU) was immediately read on the microplate reader (Molecular Devices) in the luminescence mode, reading for all wavelengths, at the endpoint read type, and analyzed with SoftMax Pro (v. 7.3.1) software.

Each microplate constituted one assay run and was analyzed independently. For neutralization curves, adalimumab was tested at 11 concentrations spanning 10 µg/mL to 0.00051 µg/mL in the presence of 3.0 ng/mL TNF-α. Plate layouts included two replicate rows of serial dilutions for the Reference adalimumab, and two replicate rows for each test article, yielding pseudo-replicates at each concentration. Negative control wells contained 3.0 ng/mL TNF-α without adalimumab. Experiments were conducted in triplicate assay runs.

Within each plate, responses from pseudo-replicate wells at the same concentration were averaged prior to model fitting. Concentrations were log₁₀₋transformed, and averaged responses were fitted to a four-parameter logistic (4PL) model in GraphPad Prism (version 10.4.2). Outliers among within-run replicates were identified using the Grubbs test (two-sided, α = 0.05). Curve quality was evaluated by the coefficient of determination (R²) values.

For TNF-α activity controls, 4PL fits of the TNF-α concentration-response curves were used to calculate the EC80 – the TNF-α concentration eliciting 80% of the maximal response) for each run; EC80 values were trended as a surrogate of TNF-α activity and general system suitability for that run. Negative control wells were prepared without the addition of TNF-α.

Parallelism Assessment and Relative Potency. Parallelism between the Reference adalimumab and test samples within the same run was assessed by comparing unconstrained and constrained 4PL models via an F-test (α = 0.05) in GraphPad Prism. In the unconstrained model, RLU vs log(mAb concentration) was fitted independently for each curve. In the constrained model, the reference and test curves were fitted simultaneously with shared Hill slope, top, and bottom asymptotes, while EC50 (or logEC50) parameters were allowed to differ. A p-value > 0.05 (α = 0.05) from the F-test indicated no statistical differences between the curves, supporting curve parallelism or similarity. Under this condition, only the EC50 (or logEC50) values were allowed to vary, enabling percent relative potency calculation as the ratio of the reference EC50 to the test sample EC50, multiplied by 100.

An 11-point, serial dilution of reference adalimumab (6 replicate curves per run) produced a sigmoidal dose-response curve (DRC) well fit by a four-parameter logistic (4PL) model (R² = 0.9839), with a Hill slope of 1.45 and EC50 of 57.55 ng/mL (Fig. 1b). The response curves encompassed ≥ 3 points on the slope and ≥ 4 points across the asymptotes. TNF-α-only wells (3 ng/mL) consistently exhibited low relative light units (RLU), aligning with the no-antibody asymptote (Fig. 1b).

To standardize neutralization, TNF-α was fixed at 3 ng/mL in all antibody-containing wells, and each run included a TNF-α control curve. The TNF-α control (log-dosed) yielded a sigmoidal, negatively sloped 4PL fit (R² = 0.9928) with decreasing luminescence at higher TNF-α (Fig. 1c). Signal-to-background (S/B) ratios were comparable for anti-TNF-α and TNF-α control curves (9.57 vs 9.55, respectively).

Specificity of the assay was demonstrated from the lack of apparent dose response curve from buffer and control wells. No DRC was observed for the buffer plus TNF-α wells (Fig. 1b), and for buffer-only wells that lacked TNF-α and anti-TNF-α (Fig. 1c).

Precision was assessed as repeatability (intra-assay) and intermediate precision (inter-assay). Intra-assay %CV, calculated from n = 3 replicate EC50 values within a plate/run, ranged from 3% to 12% across runs. Intermediate precision, assessed across 3 independent runs and 2 instruments, showed %CVs of 19% and 13%, (Table 1), indicating high between-run reproducibility.

Bioassays are central to the quality control of therapeutic proteins because they quantify biological function in a MOA–relevant context, complementing physicochemical tests that probe isolated attributes. Reporter gene assays (RGAs) are particularly well suited to TNF inhibitors, as engineered cells convert TNF-α–driven signaling into a sensitive luminescent output that integrate ligand binding, pathway modulation, and downstream response in a single measurement (Fig. 1). In this study, we established and partially validated a β-galactosidase enzyme-fragment complementation RGA using A549-TNFR1/IκB-β-gal cells to measure neutralization of soluble TNF-α by adalimumab and its biosimilars, following ICH Q2(R2) and USP < 1032>/<1034>.

The assay demonstrated an excellent working range, with 4PL fits capturing multiple informative points along the slope and asymptotes (Fig. 1b, c). Inclusion of a TNF-α control curve in every run standardized the neutralization context (fixed TNF-α at 3 ng/mL), verified reagent activity, and anchored the upper signal ceiling (TNF-α–negative wells), which can vary with cell density and run-to-run factors. Collectively, these design features support reliable quantitation across the expected potency range.

Precision performance was robust across study tiers. Repeatability (%CV) and intermediate precision (average %CV) confirmed stable EC50 estimates within and across plates and days (Table 1). Inter-instrument assessments further showed that EC50 and S/B parameters had %CV values of ≤ 20% (Table 1), underscoring the method’s portability. Such precision meets expectations for quantitative cell-based assays intended for routine use and reduces the likelihood of false trends during product life-cycle monitoring.

SST anchored day-to-day assay governance to historical performance. From n = 20 runs, the reference adalimumab EC50 distribution (mean ≈ 50.9 ng/mL) defined acceptance limits at ± 2 SD (32.6–69.2 ng/mL), complemented by criteria for RLU %CV ≤ 25%, S/B between 5.0 and 11.0, and Hill slope 1.0–1.8 (Table 2). Although the coefficient of determination (R²) was not a formal SST criterion, it was monitored as a quality indicator. Runs with low R² typically coincided with out-of-window EC50 or atypical slope and were excluded, demonstrating the value of multiple, orthogonal SST elements. This framework provides operational resilience while avoiding over-constraint that could otherwise reject biologically acceptable runs.

Robustness tests defined practical boundaries for execution. Varying the TNF-α pre-incubation duration showed that 30–45 min maintained EC50 values within SST limits and yielded consistent 4PL fits, whereas 15 min introduced excessive variability (%CVs increased in slope and S/B; Fig. 2; Supplementary Table S1). These findings support a ≥ 30-min pre-incubation as the recommended condition. Importantly, performance on an alternate plate reader also met all SST criteria (Table 1), indicating that the method tolerates reasonable differences in detection systems when governed by SST.

Further studies demonstrated the bioassay’s applicability for evaluating biosimilars. Both BS1 and BS2 produced sigmoidal dose–response curves with excellent 4PL fits (R² ≥ 0.94) and satisfied parallelism criteria relative to the reference product using constrained models (shared slope and asymptotes; F-tests p > 0.05; Fig. 3; Supplementary Table S2). The relative potencies were 103% (BS1) and 96% (BS2), each within the predefined acceptance interval for this reporter gene assay (80–137%). Although only a limited number of lots were tested, these findings support the utility of the bioassay’s suitability for biosimilar development and warrant further studies to evaluate its stability-indicating capability using stressed samples, such as those exposed to heat or light.

The platform’s MOA relevance and modular features suggest adaptability across the TNF inhibitor class. With appropriate qualification such as re-establishing SST limits, verifying parallelism, and confirming TNF-α control behavior, the same cellular backbone and readout can be extended to other TNF-targeted biologics (e.g., infliximab, etanercept, certolizumab pegol, golimumab) and their biosimilars. A unified, mechanism-relevant bioassay strategy may streamline development, facilitate bridging across manufacturing changes, and support regulatory comparability submissions.

This work also clarifies the assay’s scope and technical considerations. The readout specifically reflects neutralization of soluble TNF-α via TNFR1-linked NF-κB signaling in A549 cells and does not directly assess interactions with transmembrane TNF or off-pathway mechanisms. Because cell-based assays are inherently sensitive to biological variables such as passage number, plating density, and cytokine lot, rigorous control of cell health, inclusion of TNF-α activity controls, and continuous control charting of EC50, slope, and S/B are essential. In practice, orthogonal methods such as ligand binding or targeted physicochemical analyses, should be integrated into a comprehensive control strategy, with the RGA serving as the primary functional readout.

In summary, the qualified RGA is fit-for-purpose to quantify the functional activity of adalimumab and its biosimilars. Its demonstrated precision, robustness to operational variables, and SST-guided performance, combined with strong MOA relevance, support its use for lot release and comparability testing (Figs. 1–3; Tables 1–3). When applied alongside orthogonal analytics, this platform strengthens the quality framework for TNF-α inhibitors and enables consistent, clinically relevant potency control throughout the product life cycle.