A high-performance, non-radioactive potency assay for measuring cytotoxicity: A full substitute of the chromium-release assay targeting the regulatory-compliance objective

ABSTRACT

Standardized and biologically relevant potency assays are required by the regulatory authorities for the characterization and quality control of therapeutic antibodies. As critical mechanisms of action (MoA) of antibodies, the antibody-dependent cell-meditated cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC) must be characterized by appropriate potency assays. The current reference method for measuring cytotoxicity is the ⁵¹Cr-release method. However, radioactivity handling is difficult to implement in an industrial context because of environmental and operator protection constraints. Alternative non-radioactive methods suffer from poor validation performances and surrogate assays that measure FcγR-dependent functions do not comply with the regulatory requirement of biological relevance. Starting from these observations, we developed a non-radioactive luminescent method that is specific for target cell cytolysis. In adherent and non-adherent target cell models, the ADCC (using standardized effector cells) or CDC activities of rituximab, trastuzumab and adalimumab were compared in parallel using the ⁵¹Cr or luminescent methods. We demonstrated that the latter method is highly sensitive, with validation performances similar or better than the ⁵¹Cr method. This method also detected apoptosis following induction by a chemical agent or exposure to ultraviolet light. Moreover, it is more accurate, precise and specific than the concurrent non-radioactive calcein- and TR-FRET-based methods. The method is easy to use, versatile, standardized, biologically relevant and cost effective for measuring cytotoxicity. It is an ideal candidate for developing regulatory-compliant cytotoxicity assays for the characterization of the ADCC, CDC or apoptosis activities from the early stages of development to lot release.

KEYWORDS:

• ADCC
• apoptosis
• CDC
• cell-based assay
• cytotoxicity
• potency
• regulatory compliance
• therapeutic antibody

Abbreviations

Ab	=	antibody
ADCC	=	antibody-dependent cell-meditated cytotoxicity
CDC	=	complement-dependent cytotoxicity
CI	=	confidence interval
CV	=	coefficient of variation
E:T	=	effector:target
FBS	=	fetal bovine serum
RLU	=	relative luminescence units
S/B	=	signal/background
SD	=	standard deviation
TR-FRET	=	time-resolved fluorescence resonance energy transfer
USP	=	United States Pharmacopeial Convention

Introduction

International regulatory authorities, such as the Food and Drug Administration or European Medicines Agency, require the inclusion of a wide panel of characterization approaches in the quality control process associated with therapeutic antibody (Ab) production.^1,2 In addition to the physicochemical (e.g., sequence) and immunological (e.g., affinity, cross-reactivity) properties or purity and contaminant quantification, the biological activity must be characterized by using the appropriate in vitro potency assays. This important role of potency assays has been further strengthened by the expansion of the biosimilar market because biological activity tends to be considered a key factor in demonstrating biosimilarity.^3-5 According to the regulatory definition, the potency represents “the quantitative measure of biological activity based on an attribute of the product which is linked to the relevant biological properties.”¹ Potency assays are used to characterize the product and monitor lot-to-lot consistency and stability;² thus, highly precise, accurate and robust assays must be developed to ensure the therapeutic product reliability. Similar to all characterization and quality control purposed assays, potency assays must be validated according to very strict guidelines, such as the International Conference on Harmonization (ICH) Q2 (R1)⁶ or the United States Pharmacopeial Convention (USP) chapters <1032>, <1033> and <1034>.⁷

The induction of cytotoxicity against target antigen-expressing cells is a key mechanism of action of several therapeutic Abs, particularly in the case of anti-tumor human IgG1. This cytotoxic activity may be mediated directly through antigen binding, which leads, for example, to apoptosis induction, or to cell killing through immune effector mechanisms that involve the Fc portion of the Ab and result in the complement cascade activation (CDC) or the recruitment of Fc receptor-expressing cytotoxic effector cells (e.g., natural killer cells or macrophages) that mediate target cell lysis via ADCC or phagocytosis.⁸ Numerous publications highlight the role of the interaction between the Fc portion of IgG1 and Fcγ receptors, and particularly FcγRIIIa (CD16a) in humans, in the functional response to several therapeutic antibodies in vitro⁹ and in vivo,^10-12 assigning a central role to the ADCC in the mechanism of action of these antibodies. Furthermore, the improvement of CD16a binding by glyco- or protein-engineering has yielded Fc-optimized antibodies with higher in vivo efficacy in pre-clinical and clinical studies,^13-16 which has resulted, for example, in the recent approval of Gazyvaro® (anti-CD20, obinutuzumab).¹⁷

In this context, the availability of a relevant potency assay to measure Ab-induced cytotoxicity and, more specifically ADCC activity, is a key factor in the development of therapeutic antibodies to ensure candidate screening, production optimization and lot-to-lot consistency. Initially reported in the 1960s,^18,19 the commonly used ⁵¹Cr-release assay (similar to other radionuclide-based assays) has been considered the most sensitive and biologically relevant assay for cytotoxicity. As a result of the relatively low level of ⁵¹Cr spontaneous release by the radiolabeled cells and the high sensitivity provided by the radioactive signal, the method is sensitive and provides a good signal/background (S/B) ratio, even in the presence of a limited number of target cells per test (1,500 to 3,000 cells). These characteristics result in good performances in terms of accuracy, precision and robustness, at least for a complex bioassay, such as an ADCC assay. In addition, the underlying mechanism of ⁵¹Cr release is fully consistent with the biological phenomenon of the ADCC and is thus compliant with regulatory requirements regarding potency assays. However, taking into consideration the advances in environmental protection and operator safety, the use of radionuclides is increasingly constraining and costly and is nearly impossible to implement in an industrial context.

Several non-radioactive alternatives to the ⁵¹Cr-release assay have been described or are commercially available. These methods are based on a direct cell death measurement or an indirect measurement of a surrogate event more or less closely associated with cell death. The direct methods include target cell labeling with non-radioactive molecules, such as calcein or time-resolved fluorescence resonance energy transfer (TR-FRET) probes (e.g., lanthanide chelates), which may be detected following cell death-induced release on a principle similar to the ⁵¹Cr-release method. However, these methods are dependent on the activity of intracellular esterases (required to activate the cell-permeable pro-forms of the reporter molecules in the cytoplasm), which results in target cell line-dependent labeling variations. Moreover, both calcein and lanthanides exhibit high to very high levels of spontaneous release,^20-22 which result in low sensitivity assays compared with the ⁵¹Cr-release assay, despite the higher number of target cells required per assay (classically 5,000 to 15,000). Another group of direct and specific methods for evaluating target cell death in an ADCC assay is based on flow cytometry.^23-25 Combining differential labeling of target and effector cells with viability markers, these methods specifically measure target cell death or disappearance. However, they also suffer from 2 classical limitations of flow cytometry, low throughput and relatively high sample-to-sample variations, which result in insufficient precision and robustness for regulatory-compliant use (internal unpublished results).

A third group of methods that directly measure cell death is based on the measurement of ubiquitary and constitutively expressed enzymes or molecules ^26-28 released during the cytolytic process. Despite good performances in terms of the validation potential for evaluating the death of a homogeneous cell population, these methods are not pertinent when 2 or more cell types are co-cultured in a single well, which is, for example, the case in an ADCC assay. Effector cells that die during the cytolytic process will participate in the global enzymatic release, and it will not be possible to discriminate between the target- and effector-originating signals. A similar reasoning may be applied to impedance-based methods, in which the cell adherence is measured thanks to an impedance-sensitive chip.^29,30 In addition to being limited to the sole use of adherent target cells, this method only evaluates the growth or disappearance of cells through their adhesion to the substrate; it does not measure cell death. Overall, there is no fully satisfactory non-radioactive method for measuring target cell death in the context of a regulatory-compliant ADCC assay with good validation performances.

To circumvent these difficulties in validating ADCC assays that measure the actual target cell death, various indirect surrogate methods have been developed. In general, these surrogate assays are based on a measurement of CD16-dependent functions using FcγR-expressing reporter cell lines associated with an appropriate reporter system. For example, Promega markets a Jurkat cell line that is genetically transformed to express CD16 (for an ADCC-like assay) or CD32 (for an Ab-dependent cellular phagocytosis [ADCP]-like assay) and a luciferase reporter gene under the dependence of the FcγR-associated NFAT signaling pathway. In anti-CD20 models, the method exhibited good validation potential.³¹ Several other methods are based on a similar principle of measuring reporter effector cell activation in response to an Ab to extrapolate its ADCC activity.^32,33 Nevertheless, the main drawback of these methods is that they assume that the CD16-dependent response is predictive of the ADCC activity, and there are several reports that provide findings against this assumption.^34,35 For example, in 2009, Kawaguchi et al. demonstrated that the physiologic state or the initiation of mechanisms of resistance to the perforin/granzyme B pathway may modify the ADCC sensitivity of target cells.³⁵ These findings are likely model-dependent; however, in the absence of a clear demonstration that the CD16-dependent response and ADCC activity are correlated in a specific model, these surrogate assays, which are based on the activation of effector cells, are not compliant with the rules of regulatory authorities regarding potency assays.

Based on these different observations, we aimed to develop a non-radioactive method with the objective of regulatory compliance. The method should be specific for target cell lysis, easy to use, based on an inexpensive and widely spread readout method and should exhibit similar or better performances than the ⁵¹Cr-release assay.

Results

Selection of a reporter molecule

Our initial objective was to develop a method based on a principle similar to the ⁵¹Cr-release gold standard method, i.e., measuring an intracytoplasmic molecule that is released in the culture supernatant by the dying target cells. This molecule had to be specifically expressed by the target cells as a free form in the cytoplasm and exhibit release kinetics comparable to ⁵¹Cr release; for example, it should reach its maximum 3 to 6 hours following reaction initiation in the case of an ADCC assay. To avoid the use of an extemporaneous chemical labeling (such as calcein or ⁵¹Cr), which is a source of potential inter-assay variability, we chose to construct genetically transformed target cells that constitutively express a reporter protein.

Several reporter proteins have been tested, which vary in size and readout technology. Target cell lines were transfected with an expression vector that encoded the tested reporter protein, and the screening parameters for evaluating these molecules included the linearity of the signal, the signal/background (S/B) ratio, and the kinetics of the release. We first tested a 47 kDa chimerical protein, which consisted of 2 tag sequences and was detected with specific antibodies through a lanthanide-based TR-FRET readout. Despite a bright signal following cell lysis with Triton X-100, which resulted in a good S/B ratio (> 15), no protein release was identified in a classical ADCC set up after 4 hours of incubation. Moreover, at least 20 hours were required to identify significant release, which led to highly variable results (data not shown).

We subsequently decided to test the nanoluciferase protein. Nanoluciferase is a recently described ³⁶ protein of 19 kDa derived from the catalytic subunit of the deep-sea shrimp Oplophorus gracilirostris. Two target cell models were constructed to stably express nanoluciferase, the non-adherent Raji B cell lymphoma and the adherent SKOV-3 ovarian cancer cell lines. Preliminary data using nanoluciferase-containing cell culture supernatants demonstrated that the signal was linear between ∼3,000 and 1,000,000 relative luminescence units (RLU), which enabled a wide range analysis of 2.5-Log₁₀. We subsequently assessed the efficiency and sensitivity of the method in both cellular models in the presence of decreasing amounts of nanoluciferase-expressing target cells by evaluating both the linearity of the spontaneous and maximum release signals and the resulting S/B ratio in typical ADCC assay conditions. We used 3,000, 1,500, 1,000 and 500 cells per well, with or without lysis with Triton X-100, and the results are shown in Fig. 1. The data initially confirmed the linearity of the signal in the previously defined range and a typical ADCC assay experimental setting. For Raji cells, the maximum signal did not exceed 1.5×10⁵ RLU, and the linear determination coefficients r² were 0.9731 and 1.0000 for the spontaneous and maximum release signals, respectively (Fig. 1A). The resulting S/B ratios were approximately constant, regardless of the amount of cells, which ranged from 7.0 to 9.4 (Fig. 1B). For the SKOV-3 cells, the maximum release signal ranged from 2.7×10⁵ to 1.0×10⁶ RLU for 500 and 3,000 cells per well, respectively (Fig. 1C). The coefficients r² were 0.9936 and 0.9949 for the spontaneous and maximum release signals, respectively. The resulting S/B ratios ranged from 14.5 using 500 cells/well to 7.8 using 3,000 cells (Fig. 1D).

Figure 1. Characterization of the spontaneous and maximum release signals of nanoluciferase-expressing cells. Varying numbers of nanoluciferase-expressing Raji (A and B) or SKOV-3 (C and D) cells were incubated in triplicates for 4 h at 37°C in RPMI-1640 with 5% FBS that contained an irrelevant monoclonal Ab (trastuzumab for the Raji cells and rituximab for the SKOV-3 cells) and ADCC effector cells (E:T = 10:1), in the presence or absence of Triton X-100 to measure the maximum and spontaneous release signals, respectively, in a typical ADCC experimental setting. The nanoluciferase signal was measured as described in the Materials and Methods section. Left graphs (A and C) represent the measured spontaneous (open circles) or maximum (black squares) signals released by the different amounts of cells (mean ± SD). Linear regressions were performed for each set of data and are shown in the left graphs (dotted line for spontaneous and solid line for maximum release). For each evaluated amount of cells per well, the S/B ratio was calculated as the mean of the maximum signal divided by the mean of the spontaneous release and is shown in the right graphs (B and D).

These data demonstrated that very good S/B ratios were achieved with the nanoluciferase method, even with a very low number of target cells. Consistently, very satisfactory spontaneous release percentages were demonstrated with nanoluciferase, thereby reaching 10% and 13% for the Raji and SKOV-3 cells, respectively, regardless of the amount of cells. For comparison, our data exhibited 18% and 23%, respectively, in the same conditions using ⁵¹Cr labeling.

Assessment of nanoluciferase release as a consistent marker of cell death in several cytotoxicity models

The next step was to demonstrate that the nanoluciferase release was correlated with actual cell death through a physiologic mechanism. Thus, we induced apoptosis in the nanoluciferase-expressing Raji clone, with increasing doses of staurosporine or increasing durations of ultraviolet (UV) light exposure. After 48 h of culture, the nanoluciferase signal was measured in the supernatants, and the viability of the cells was measured via flow cytometry in the corresponding pellets. The percentages of dead cells were calculated from both methods, and the results are shown in Fig. 2. A very good correlation was found between both methods, with a regression slope approximately equal to 1 (0.97 for both UV and staurosporine induction), regardless of the type of apoptosis induction, i.e., physical or chemical. This finding demonstrated that the nanoluciferase release method was significantly representative of the target cell death and directly proportional to the actual dead cell percentage.

Figure 2. Correlation between cell death and nanoluciferase release. 5.0×10⁴ nanoluciferase-expressing Raji cells/well were cultured for 48 h in RPMI-1640 with 10% FBS [A] in the presence of increasing doses of the apoptosis-inducing drug staurosporine (6,4 nM to 20 µM) or [B] after increasing periods of UV exposure (10 to 60 s). In each well, the percentage of dead cells was calculated by measuring the nanoluciferase signal in the supernatant and via flow cytometry on the cell pellets, as described in the Materials and Methods section. For each tested condition in 2 independent assays, the graphs summarize the percentage of dead cells based on the nanoluciferasedetermination (x-axis) vs. the percentage of dead cells based on the flow cytometry method (y-axis). The dotted line indicates the linear regression calculated from the data. The regression equations were y = 0.97x – 2.75 (r² = 0.9830) for staurosporine-induction (A) and y = 0.97x – 4.90 (r² = 0.9595) for UV induction (B).

We subsequently assessed the kinetics of the nanoluciferase release in ADCC and CDC assays, particularly in comparison to the ⁵¹Cr-release method, which are classically performed during 3 to 4 hours. The ADCC and CDC assays were thus performed using increasing incubation times of 1, 2, 3 or 4 hours using Raji cells (Fig. 3A and B). The SKOV-3 cell line, similar to many carcinoma-derived cell lines, strongly expressed complement resistance factors ³⁷ and was thus non-sensitive to the CDC; thus, only the ADCC activity was tested with this model (Fig. 3C). For the ADCC assays, the nanoluciferase release was time-dependent, with an activity plateau reached after 3 hours in both models. For the CDC, the maximum lysis was obtained after 2 hours. Overall, these assay durations were completely similar to the classical ⁵¹Cr-release based method.

Figure 3. Kinetics of nanoluciferase release in ADCC and CDC assays. Nanoluciferase-expressing non-adherent Raji (A and B) or adherent SKOV-3 (C) cells were used in the ADCC (A and C) or CDC (B) assays performed during 1, 2, 3 or 4 h. At the end of the incubation, supernatants were collected, and the nanoluciferase activity was measured. The graphs indicate the percentage of the resulting specific lysis (mean of triplicates, y-axis) for each Ab concentration (Log₁₀-transformed, x-axis) and the corresponding 4-parameter logistic regression models. The data were obtained from a single experiment, with each condition performed in triplicate.

Moreover, an important consideration to ensure the reproducibility of the cytotoxicity experiments was the stability of the nanoluciferase expression. Hence, this stability was assessed with both the Raji and SKOV-3 selected clones during 50 passages, and the results are shown in Fig. 4. Three thousand cells were used for each determination, and the average RLU values (mean ± SD, coefficient of variation [CV] between brackets) along the 50 passages were 125,674 ± 36,505 RLU (29.1%) for the Raji clone (n = 20 determinations) and 842,262 ± 190,569 RLU (22.6%) for the SKOV-3 clone (n = 26 determinations). Taking into account that the experimental process involved a manual counting step (with 20% potential variation), a variability of 40% of the median RLU value was considered acceptable. Consequently, with the exception of few outliers, the stability of the nanoluciferase expression was considered satisfactory.

Figure 4. Nanoluciferase expression stability. The selected nanoluciferase-expressing Raji (left graph) and SKOV-3 (right graph) clones were monitored during 50 passages for nanoluciferase expression. Three thousand cells were lysed with Triton X-100, and the resulting luminescence signal was measured following substrate addition as described in the Materials and Methods section. Each dot represents an individual determination, and the median ± 40% is shown as large and small dotted lines, respectively.

Comparison of nanoluciferase- and ⁵¹Cr-release methods in ADCC and CDC assays

The nanoluciferase-release method was then directly compared with the ⁵¹Cr-release method in ADCC and CDC assays. To improve the ADCC assay reproducibility, a human CD16-transduced cytotoxic T lymphocyte clone was used following standardized production.³⁸ As a non-adherent target cell model, nanoluciferase-expressing Raji cells were used in both the ADCC and CDC experiments. Because of their resistance to CDC, the SKOV-3 adherent cell model could only be used for the ADCC assay. Thus, as an adherent target cell model for the CDC assay, we selected an internally developed Chinese hamster ovary (CHO) cell line that expressed a low-cleavage mutant form of membrane tumor necrosis factor (TNF) (CHO/TNF, unpublished data), which has also been transformed to stably express nanoluciferase. The average nanoluciferase expression of the CHO/TNF clone was similar to the previously described SKOV-3 clone (824,566 ± 162,105 RLU).

Three independent ADCC or CDC experiments were performed. For each target cell model, nanoluciferase-expressing cells were or were not labeled with ⁵¹Cr, and both were run in parallel in the ADCC assay (using rituximab for Raji cells and trastuzumab for SKOV-3 cells) or in the CDC assay (using rituximab for Raji cells and adalimumab for CHO/TNF cells). In each assay, Ab preparations, effector cell suspensions and complement dilutions were shared between the ⁵¹Cr-labeled and unlabeled target cells. After 4 h of incubation, the specific lysis was measured using a ⁵¹Cr- or nanoluciferase-release readout. The results are shown in Fig. 5, and the model characteristics are summarized in Table 1. Each condition was tested in triplicates and average intra-replicate variations were 3.8% and 7.3% for ⁵¹Cr- and nanoluciferase-based methods, respectively. For both assays (ADCC and CDC) and both models (adherent and non-adherent target cells), the nanoluciferase-based method provided very similar results to the ⁵¹Cr-based method. A deeper analysis of the modeling parameters (Table 1) indicated nearly identical E_min and E_max values between both methods for these 2 models. The EC₅₀ were also very comparable between both methods; however, a significantly lower value was identified with the nanoluciferase method (p  =  0.0093, paired non-parametric Wilcoxon's test, n  =  12), with a ∼2-fold factor, with the exception of the CHO/TNF CDC model. An analysis of the EC₅₀ CVs also confirmed the similarity of both methods in terms of the inter-assay variability, with a slight, but not significant (paired t test), advantage for the nanoluciferase-based method, which generally exhibited smaller CVs (7.2% to 22.8% versus 18.6% to 38.4%, respectively).

Figure 5. Direct comparison between ⁵¹Cr- and nanoluciferase-based methods. ADCC assays (A and B) were performed using nanoluciferase-expressing Raji cells (A) as a non-adherent target cell model and SKOV-3 cells (B) as an adherent target cell model. The CDC assays (C and D) were performed using nanoluciferase-expressing Raji (C) and CHO/TNF cells (D) as non-adherent and adherent target cell models, respectively. Cells were or were not labeled with ⁵¹Cr and were subsequently run in parallel in the ADCC or CDC assays. The percentages of specific lysis (y-axis) obtained in the presence of the indicated Ab concentrations (Log₁₀-transformed, x-axis) are indicated on the graphs, with the ⁵¹Cr-release method (open squares) or the nanoluciferase-release method (black circles), together with the calculated unconstrained 4-parameter logistic regression models (dotted and solid lines, respectively). The data were obtained from 3 independent assays (mean ± SD).

Table 1. Comparison of the characteristic parameters of ADCC and CDC assays using ⁵¹Cr and nanoluciferase methods in the unconstrained 4-parameter logistic regression model. Arithmetic mean values and ranges (minimum/maximum) of the EC₅₀ (in ng/mL), E_min and E_max (both in % of specific lysis), as well as the coefficient of variation (CV) of the EC₅₀ are indicated in the table for both the ADCC and CDC assays using ⁵¹Cr or nanoluciferase readouts. The data were obtained from 3 independent experiments.

Assay	Model	Method	EC50 (ng/mL) [min / max]	CV of EC50 (%)	Emin (%) [min / max]	Emax (%) [min / max]
ADCC	Raji	^51Cr	55.7 [39.3 / 72.0]	29.3	−1.2 [−2.8 / −0.2]	75.7 [69.3 / 81.2]
		Nanoluciferase	31.8 [23.7 / 37.8]	22.8	−1.2 [−2.5 / −0.1]	73.6 [66.7 / 83.0]
	SKOV-3	^51Cr	387.7 [247.8 / 544.1]	38.4	0.4 [−1.7 / 2.9]	41.2 [37.6 / 39.9]
		Nanoluciferase	191.6 [178.4 / 205.8]	7.2	0.1 [−0.4 / 0.4]	35.1 [31.5 / 42.1]
CDC	Raji	^51Cr	156.7 [134.7 / 191.0]	19.2	3.0 [0.9 / 7.0]	88.6 [79.9 / 94.8]
		Nanoluciferase	73.1 [65.6 / 84.3]	13.5	−0.3 [−3.8 / 2.5]	95.6 [88.3 / 102.3]
	CHO/TNF	^51Cr	142.7 [112.9 / 164.1]	18.6	−0.5 [−1.9 / 1.6]	47.9 [45.5 / 50.9]
		Nanoluciferase	152.5 [131.0 / 176.4]	15.0	−2.2 [−4.1 / −0.9]	46.5 [42.0 / 50.0]

Performance of the nanoluciferase assay compared with the ⁵¹Cr-release assay

A cytotoxicity assay compliant with the regulatory guidelines on monoclonal Abs must follow international guidelines regarding data analysis and validation of the performance criteria, such as the accuracy, precision or robustness. Accordingly, we designed a validation procedure based on both the ICH Q2(R1) guidelines ⁶ and the recently published USP chapters <1032>, <1033> and <1034> regarding Bioassay Validation. In addition to the standard Ab dilution ranges described in the previous paragraph and Fig. 5, which is considered the “reference sample”, each reference Ab was diluted or concentrated at 80% or 125% of the nominal concentration to yield under- and over-potent samples (which mimicked batch-to-batch variations from the same mAb). These “test samples” were subsequently run in parallel with the reference sample in both the ⁵¹Cr- and nanoluciferase-based ADCC and CDC assays, using nanoluciferase-expressing target cells in both methods. To follow a worst-case approach and obtain a preliminary idea of an assay's robustness, the 3 assays were performed by 2 different operators on 2 different days for each model.

The amount of data available to us was too small to calculate equivalence margins as recommended by the USP equivalence testing approach, which prevented the calculation of a “relative potency” in the strict regulatory meaning. Thus, we based our analysis on the EC₅₀ ratio between the reference and test samples in the unconstrained model, as a surrogate of a strict relative potency calculation. For each test sample, the bias (i.e., the ratio between the measured and expected “potency”) was subsequently calculated. This analysis was performed in each model for the 3 assays in both methods, and the results are summarized in Table 2.

Table 2. Summary of EC₅₀ ratios and resulting biases calculated for under- (80%) and over- (125%) potent samples in the different ADCC and CDC models. The ratio between the reference and the under- or over-potent samples was calculated for each assay, and the table indicates the arithmetic mean values and CV. The biases represent the difference between the calculated and expected (0.8 and 1.25 for 80% and 125% samples, respectively) ratios. The table indicates the biases' arithmetic mean values and ranges (minimum/maximum). The data were obtained from 3 independent experiments.

Assay	Model	Method	EC50 ratio @ 80% [CV%]	Bias @ 80% (%) [min / max]	EC50 ratio @ 125% [CV%]	Bias @ 125% (%) [min / max]
ADCC	Raji	^51Cr	0.77 [14.2%]	−4.2% [−19.9/4.3]	1.32 [9.0%]	5.5% [−1.2/16.4]
		Nanoluciferase	0.85 [26.3%]	6.5% [−25.7/25.7]	1.18 [21.2%]	−5.6% [−25.9/14.0]
	SKOV-3	^51Cr	0.83 [15.7%]	3.6% [−15.0/14.9]	1.54 [16.1%]	23.4% [0.5/35.4]
		Nanoluciferase	0.69 [10.7%]	−13.4% [−20.2/−2.8]	1.35 [14.9%]	7.7% [−3.1/26.1]
CDC	Raji	^51Cr	0.70 [17.6%]	−13.0% [−29.1/1.5]	1.32 [12.4%]	5.2% [−5.8/19.7]
		Nanoluciferase	0.62 [24.6%]	−22.8% [−35.1/−0.9]	1.10 [8.6%]	−11.9 [−20.7/−7.0]
	CHO/TNF	^51Cr	0.87 [33.8%]	9.4% [−31.5/40.6]	1.42 [35.7%]	13.3% [−33.4/37.0]
		Nanoluciferase	0.85 [15.6%]	6.6% [−12.1/19.9]	1.31 [35.9%]	5.1% [−19.0/48.6]

Globally, both methods exhibited a similar potential to potency calculation, despite small differences within each model. For the ADCC assay with Raji cells, the EC₅₀ ratios and biases were comparable between the ⁵¹Cr and nanoluciferase methods (∼5%); however, a higher CV was identified with nanoluciferase (26% and 21% for the 80% and 125% test samples, respectively, compared with 14% and 9% for ⁵¹Cr). In the adherent cell ADCC model (SKOV-3), ⁵¹Cr had slightly higher CVs (15.7% and 16.1% for both test samples, respectively, compared with 10.7% and 14.9% for nanoluciferase), a lower bias for the 80% sample (3.6% vs. −13.4%, respectively) and a higher bias for the 125% sample (23.4% vs. 7.7%, respectively). The same types of small variations were identified in the CDC models, in which ⁵¹Cr globally exhibited slightly better performances in the non-adherent cell model, but slightly worse performances in the adherent cell model. Overall, both methods appeared to provide very similar results, which confirmed the comparability of the release mechanisms identified in the previous figures between ⁵¹Cr and nanoluciferase.

Comparison of the nanoluciferase assay to the other non-radioactive methods calcein and TR-FRET

We finally compared the nanoluciferase assay to 2 other non-radioactive methods already available for cytotoxicity measurement, the calcein release method and the DELFIA® (from Perkin Elmer) TR-FRET method. The principle of these methods is to label the target cells with acetoxymethyl (AM) pro-forms of calcein or of the fluorescence enhancer terpyridine-6,6″-dicarboxylate (TDA). The AM moiety is then cleaved by non-specific intracellular acetyl esterases, decreasing the membrane permeability of the molecules. At the end of the assay, the presence of free calcein or TDA in the supernatants is detected by direct fluorescence for calcein or by TR-FRET signal after addition of an europium solution resulting in the EuTDA chelate formation, respectively.

In this study, the nanoluciferase-expressing Raji and SKOV-3 cells were labeled with calcein or TDA, and ADCC was measured in simultaneous experiments using in parallel nanoluciferase-, calcein- or TDA-release read-outs. As described previously, a reference sample (100%) as well as under- (80%) and over- (125%) concentrated test samples were tested in each assay. With Raji cells, low to very low S/B ratios were obtained with calcein and TR-FRET methods compared with nanoluciferase (1.43 ± 0.27 and 1.15 ± 0.13 vs. 7.82 ± 1.03, respectively). These low ratios resulted in poorly reproducible results along the 3 independent experiments, as illustrated by the strong variations of the E_min, E_max and EC₅₀ compared with the nanoluciferase parameters, and consistently with low correlation factors r² (Table 3 and Fig. 6A). Average intra-replicate variations were good for TR-FRET (3.9%) and nanoluciferase (7.8%) methods, and a bit higher for calcein (16.4%). Furthermore, the apparent maximum lysis was completely aberrant, reaching more than 300% of specific lysis in some experiments. Despite the strong inter-assay asymptote variations with the calcein and TR-FRET methods, experimental biases were also calculated from the test samples and they consistently showed a poor to very poor accuracy of these 2 methods.

Table 3. Comparison of the characteristic parameters of ADCC assays using the nanoluciferase, calcein or TR-FRET methods in the unconstrained 4-parameter logistic regression model. The EC₅₀ (means and the corresponding CV), E_min and E_max (mean and range) from the reference sample are indicated in the table for both cellular models, using the different readouts. Correlation coefficients (r²) are also indicated to illustrate the data fitting to the regression model. The relative biases (and ranges) obtained with the under- and over-concentrated test samples (80% and 125%) are also shown. The data were obtained from 3 independent experiments (see also Fig. 6A). * In the Raji/calcein model @125%, only one experiment over 3 gave data fitting with the logistic model.

		Reference sample (100%)				Under or over-concentrated test samples
Model	Method	EC50 (ng/mL) [%CV]	r2	Emin (%) [min / max]	Emax (%) [min / max]	Bias @ 80% (%) [min / max]	Bias @ 125% (%) [min / max]
Raji	Nanoluciferase	14.0 [58.4%]	0.9715	−2.7 [−5.2 / −0.6]	95.9 [93.2 / 100.0]	−5.5 [−25.1 / 7.5]	16.0 [−23.9 / 58.8]
	Calcein	53.1 [144.4%]	0.7696	−37.1 [−102.7 / 23.6]	194.1 [99.1 / 376]	532.1 [−69.1 / 1689.7]	−62.7 [NA]*
	TR-FRET	13.2 [25.9%]	0.5672	−38.16 [−57.1 / −7.4]	166.5 [80.1 / 336.6]	188.3 [−100.0 / 762.7]	−32.0 [−265.0 / 160.5]
SKOV-3	Nanoluciferase	46.6 [16.5%]	0.9753	−0.5 [−1.6 / 0.5]	52.5 [45.3 / 59.7]	−26.0 [−40.0 / −11.1]	−3.2 [−30.4 / 24.9]
	Calcein	48.51 [11.4%]	0.9113	−6.3 [−20.1 / 8.6]	93.2 [78.1 / 103.6]	50.5 [6.9 / 123.5]	14.8 [−5.2 / 38.3]
	TR-FRET	25.5 [12.1%]	0.9684	0.33 [−2.8 / 4.1]	90.6 [86.6 / 94.1]	−45.2 [−50.5 / −35.2]	19.3 [1.0 / 38.0]

Figure 6. Direct comparison of the nanoluciferase method to the calcein and TR-FRET methods. Nanoluciferase –expressing target cells (Raji or SKOV-3) were used to measure ADCC activities either directly by nanoluciferase-release read-out or after calcein or BATDA labeling. (A) The resulting percentages of specific lysis and unconstrained 4-parameter logistic regressions are shown for the reference Ab (100%, black circle and plain line) and for the under-concentrated (80%, open triangle and short-dotted line) and over-concentrated (125%, open square and long-dotted line) test samples. The upper and lower panels show the results obtained with the Raji cells (note the wide Y-axis range variations between the different methods) and with the SKOV-3 model, respectively, using the 3 methods (nanoluciferase on the left, calcein on the center, and TR-FRET on the right). Data are the means ± SD from 3 independent experiments. (B) After supernatant harvesting for nanoluciferase, calcein and TR-FRET signal measurements, SKOV-3 cell pellets were washed and stained for measuring the percentage of remaining living target cells by flow cytometry (see the Material and methods section for the analysis details). The resulting percentages are shown on each dot plot. Only the spontaneous release and the lowest and highest antibody concentrations are depicted on the figure. Data are from one experiment representative of 2. (C) The E_max parameters (from unconstrained 4-parameter logistic regressions) obtained in each assay (nanoluciferase, calcein or TR-FRET) using both read-out, i.e., the respective supernatant (SN) read-out (light gray bars) and the flow cytometry (FC) measurement on the pellets (black bars), are shown. Data are mean ± SD from 2 (flow cytometry) or 3 (supernatant reading) experiments. Using the nanoluciferase supernatant condition as a reference, Student's t tests were performed against the other conditions and results are shown on the graph: NS, not significant (p > 0.05); * p ≤ 0.05; ** p ≤ 0.005.

With the SKOV-3 cells, notwithstanding the low S/B ratios obtained with the calcein- and TR-FRET methods (1.63 ± 0.16% and 1.93 ± 0.05%, respectively, compared with 21.80 ± 2.59% with the nanoluciferase read-out), the results appeared more reproducible than in the Raji model, with global performances relatively close to the nanoluciferase method (see Table 3 and Fig. 6A). Average intra-replicate variations were 4.5%, 2.4% and 8.2% for calcein, TR-FRET and nanoluciferase methods, respectively. Despite a comparable inter-assay precision between the 3 methods (illustrated by the EC₅₀ coefficients of variation in the Table 3), the nanoluciferase assay tends to exhibit a better accuracy (i.e., lower biases, in average) (Table 3). This last point was consistent with the better asymptote and slope similarities in the nanoluciferase assays between the reference and the test samples (and among the 3 experiments), compared with the calcein and TR-FRET methods. Indeed, for the nanoluciferase, calcein and TR-FRET methods, the maximum E_max CV were 6.4%, 23.5% and 12.9%, respectively, and maximum slope CV were 18.2%, 43.4% and 44.9%, respectively. More surprisingly, a major difference appeared between the methods in this model, since the SKOV-3 maximum lysis measured using the calcein or TR-FRET methods was found to be significantly higher (90–93% in average) than the maximum lysis measured using the nanoluciferase and ⁵¹Cr methods (35–53% in average, Fig. 5, Fig. 6A, Table 1 and Table 3).

To determine the actual SKOV-3 target cell death percentage in these ADCC assay conditions, a flow cytometry-based approach was used, following a similar rationale than the data presented in the Fig. 2. At the end of the ADCC incubation period and after the supernatant harvesting for nanoluciferase, calcein or TR-FRET signal measurement, the cell pellets from the trastuzumab-based assay were resuspended, labeled with fluorescent anti-epidermal growth factor receptor (EGFR) antibody and analyzed by flow cytometry to measure the frequency of the residual living target cells (Fig. 6B). The results, summarized in Fig. 6C, showed that the specific lysis measured using the supernatant in the nanoluciferase assay (52.5 ± 2.9%, mean ± SD, n = 3) was consistent with the percentage of lysis measured by flow cytometry in both the nanoluciferase assay (46.6 ± 7.2%, mean ± SD, n = 2) and the TR-FRET assay (52.9 ± 19.2%, mean ± SD, n = 2) (p>0.05, Student t test). Hence, the maximum specific lysis measured in the supernatant of the TR-FRET assay appeared to be over-estimated (p = 0.0026 vs. nanoluciferase cytometry and p = 0.0372 vs. TR-FRET cytometry). Interestingly, in the calcein assay, both the supernatant signal measurement and the flow cytometry read-out showed a significantly increased specific lysis compared with the nanoluciferase assay supernatant measurement (93.2 ± 13.4%, p = 0.0200 and 76.7 ± 1.2%, p = 0.0222, respectively, mean ± SD and Student t test p value), while no significant difference could be found between these 2 read-outs of the calcein assay. This last result suggested an increased mortality specifically related to the calcein assay, competing with the ADCC-induced lysis. Overall, the nanoluciferase assay appeared as the most representative method for target cell cytotoxicity measurement, giving results consistent with flow cytometry and ⁵¹Cr-release methods.

Discussion

The availability of a biologically relevant and standardized bioassay technology for evaluating the cytotoxicity of therapeutic Abs and, more specifically, the ADCC activity is a key challenge in fulfilling regulatory authority requirements. To comply with the biological relevance objective, the technology should be based on an actual measurement of target cell death. To date, the most sensitive methods to evaluate target cell death are based on ⁵¹Cr (or other radionuclides, such as ¹¹¹In) release measurement. However, the use and development of potency assays based on these methods, particularly in an industrial context, are strongly hampered by the constraints of radioactive element use.

Thus, we developed a non-radioactive method to measure cytotoxicity based on the transformation of target cells to stably express a reporter protein, whose release in the culture supernatant is correlated with the target cell death. Although this approach, which involves cell transformation and clone selection steps, may initially appear to be a fastidious and long process, it makes sense in the context of therapeutic Ab development and quality control. The few weeks required to establish a stable target cell clone are largely compensated by the substantially improved stability and reproducibility of the assay over the entire lifetime of the product, which may be 15 to 20 years, provided that careful characterization is performed, coupled with Master and Working cell bank-based management to optimize the assay stability over time. Moreover, with a judicious choice of the target cell line model, one transformed clone may be used for several candidate targets or antibodies, which increases the profitability of the initial investment.

In contrast to the previously published, and partially unsuccessful, approaches that involve firefly luciferase or β-galactosidase-expressing target cells,³⁹ we selected the recently marketed nanoluciferase³⁶ as the reporter protein. Nanoluciferase has been used as a reporter gene for protein-ligand interactions and signaling pathway studies^40,41 or has been coupled to a protein of interest for live imaging^42,43 and trafficking^44,45 studies. With a substantially increased half-life in extracellular medium compared with the firefly luciferase (7.7 d vs. 30 minutes, respectively),^36,39 nanoluciferase was compatible with the typical cytotoxicity assay durations. Furthermore, as a result of its high luminescence activity, the nanoluciferase-based assay may be performed using a very low number of target cells (500 cells/well). In contrast to β-galactosidase, nanoluciferase also appeared to be quickly and completely released during cell death, as illustrated by the perfect correlation with the flow cytometry cell death measurement in the apoptosis assays (Fig. 2) and the fully comparable percentages of specific lysis with the ⁵¹Cr-release assay (Fig. 5, Table 1 and Table 2). Moreover, Schäffer et al.³⁹ identified significant differences between the percentages of specific lysis when the readout was ⁵¹Cr or β-galactosidase, and the latter always resulted in lower percentages. The mechanisms of ADCC and pore formation on the target cell membrane are complex and largely unresolved; however, our results suggest that the size of the reporter molecule matters. Lopez and colleagues⁴⁶ recently demonstrated that small pores (120 to 170 Å) were transiently formed (during 20 to 80 seconds) in the immunological synapse to enable granzyme entry into the target cells. Since the preservation of membrane integrity during granzyme-induced apoptosis remains questionable,⁴⁷ these pores may also constitute an exit gate for the intracellular content during the ADCC process. Regardless of the involved release process, nanoluciferase, with its relatively small size (19 kDa), appears to be more easily released during the cytotoxicity process compared with bigger proteins, such as β-galactosidase (464 kDa) or the initially tested 47 kDa chimerical TR-FRET reporter protein.

This finding was further supported by our investigation of the release kinetics, which demonstrated a fast release of nanoluciferase, as quick as 2 h in the CDC model and 3 h for ADCC. This quicker release in the CDC assay was consistent with its mechanism of action, in which the lytic system was constituted by soluble factors available in excess, thereby leading to concomitant lysis, in contrast to the ADCC assay, in which the effector cells had to sequentially initiate cell-to-cell contacts before killing on a “first-come, first-served” basis. Consistently, the S/B ratios observed with the nanoluciferase method (between 7 and 15, depending on the assays and models) were comparable to classic S/B ratios observed with the ⁵¹Cr-release method (between 5 and 12 in similar conditions). Furthermore, the release kinetics and characteristics are comparable between both methods. When the ADCC and CDC assays were run in parallel in a first series of experiments comparing the nanoluciferase and ⁵¹Cr-release methods, using the same nanoluciferase-expressing target cells in the same experimental settings, the similarity between the 2 methods was confirmed, which resulted in highly similar characteristic curves and modeling parameters. Using the unconstrained model (Table 1), the EC₅₀ CVs of the standard curves were relatively classical for these types of bioassays. Indeed, the majority of the assays exhibited between 15 and 25% CV variation, with the exception of 29% and 38% for the rituximab and trastuzumab ⁵¹Cr-based ADCC assays, respectively. The nanoluciferase assay performed generally better than ⁵¹Cr, with CVs that ranged from 7.2% to 22.8% and 18.6% to 38.4%, respectively.

Notwithstanding the impressive similarity between the ⁵¹Cr and nanoluciferase lysis measurements illustrated in Fig. 5 and Table 1, a slight difference was identified in the rituximab (ADCC and CDC) and trastuzumab (ADCC) models regarding the EC₅₀value, which was approximately 2-fold lower using the nanoluciferase readout compared with ⁵¹Cr. This result could not be explained by a difference in the ADCC activity of the sample because the same Ab preparation was used in each assay for both the ⁵¹Cr-labeled and non-labeled target cells. Thus, these findings suggested that despite comparable release characteristics, small differences may remain between the nanoluciferase and ⁵¹Cr release mechanisms in these models. Nevertheless, this small difference in the absolute EC₅₀ values was not considered critical because the ⁵¹Cr- and nanoluciferase-determined EC₅₀ were not supposed to be directly compared with each other, especially in a potency measurement-based approach.

Following the proof of concept that the nanoluciferase-release method was a valuable tool to efficiently measure cell death in a variety of experimental designs (e.g., apoptosis, ADCC and CDC assays), we attempted to assess its validation potential in a regulatory environment. According to the USP chapters <1032> and <1034>, the potency of a test sample may be measured relative to a reference sample by calculating their EC₅₀ ratios in a constrained 4-parameter logistic model, provided that both curves in the unconstrained model are similar, i.e., have equivalent asymptotes and slopes (parallelism assumption). For each parameter, during assay development, the USP recommends determining equivalence margins representative of the assay variability by compiling historical data. This may be achieved, for example, by running the reference vs. itself, or vs. modified or altered batches whose potency has been demonstrated to be acceptable, for approximately 25 iterations.^48,49 However, the objective of our study was not to develop a specific potency assay for a given Ab product; rather, it was to provide preliminary experiments that illustrate the potentialities of the nanoluciferase cytotoxicity based method. Thus, we did not have access to a sufficient amount of historical data to accurately mark out the acceptable variability limits of asymptotes and slopes. Because of our limited set of data, we subsequently chose to diverge from the strict “relative potency” calculation described in the USP chapters and to use the EC₅₀ ratio between the reference and standard samples from the unconstrained model as a surrogate evaluation of the relative potency. The unconstrained model was selected instead of the constrained model because, in general, constraining the model decreases the curve fitting to the experimental data points, which thus decreases the EC₅₀ data representativeness. Using under- (80%) and over- (125%) concentrated preparations of the reference sample as test samples, we applied this analysis strategy to determine the accuracy and precision of the assay. To further provide robustness elements in this evaluation of performances, 2 different operators were involved in the 3 performed assays, and the assays were performed independently on separate days.

The EC₅₀ ratios calculated for the test samples exhibited CVs that ranged from 9% to 36% (Table 2) depending on the different conditions and models, consistent with the EC₅₀ CVs calculated for the standard curves alone (Table 1), which ranged from 15% to 25%. Moreover, in general, the average experimental biases were inferior to 14%, with 2 exceptions at 23% (125% test sample in the SKOV-3 ADCC and 80% test sample in the Raji CDC). These values are classically observed or reported for these types of complex bioassays; however, they remain relatively high and may be improved in the future. This may be achieved by standardizing and optimizing the assay one step further, after identifying the criteria that critically affect the assay variability. For example, the preparation of target cells may be optimized by an assessment of the effect of several factors, such as the cell confluence before plating, the plating concentration and the time before effector cells or complement addition (e.g., to enable adherent cells to re-adhere). Even if a standardized effector cell population specifically developed to improve the ADCC assay reproducibility has been used in these experiments, optimization may likely be made on this part, for example, by adapting the effector:target (E:T) ratio to a specific target cell model.

We are conscious that the EC₅₀ ratio-based analysis selected here does not follow the USP equivalence testing approach. In a “real life” assay development, equivalence margins will be defined on the basis of a substantial amount of historical data, which may enable the determination of accurate acceptance criteria, which are representative of the true assay variability. However, importantly, the nanoluciferase-based assay did not underperform compared with the ⁵¹Cr method and exhibited at least comparable, and sometimes better, performances compared with the historical method. In some models or conditions, ⁵¹Cr had a small advantage in terms of the CV or range; however, the nanoluciferase method performed better than ⁵¹Cr in other conditions. Thus, the validation potential of the nanoluciferase method appeared to be similar to the historical ⁵¹Cr method; moreover, it had substantially increased potentials regarding developability, particularly as a result of the absence of radioactivity, and long-term reproducibility, which resulted from the stably transformed target cell approach.

The last part of our study aimed to compare the new nanoluciferase method to some alternative non-radioactive methods for cytotoxicity measurement, i.e., the calcein-release method and the TR-FRET read-out (using the DELFIA® cytotoxicity kit from Perkin Elmer). Both the nanoluciferase-expressing Raji and SKOV-3 cell lines were tested for ADCC measurement after calcein or TDA-labeling, in parallel to the nanoluciferase release assay. The data suggested that neither the calcein nor the TDA labeling was adapted to the Raji cell model, the extremely low S/B ratio resulting in inconsistent and highly variable results. This might be related to a very high spontaneous release of calcein or TDA by the Raji cells, which is a common drawback of these 2 methods.^20,21 In contrast, the SKOV-3 cell line showed relatively good accuracy and precision performances that were very similar to the nanoluciferase method. However, despite these good performances, a major difference regarding the maximum lysis evaluation by the calcein and TR-FRET methods was observed. In all our previous experiments using ⁵¹Cr or nanoluciferase read-out, trastuzumab induced a partial lysis of the SKOV-3 target cells, rarely exceeding 50 to 60%, also consistent with some previously published data.⁵⁰ In contrast, maximum lysis appeared close to 100% with both the calcein and the TR-FRET methods.

We thus explored the residual percentage of living SKOV-3 cells at the end of the ADCC assay, using a flow cytometry approach targeting EGFR, another specific receptor of the SKOV-3 cells in this model. For the nanoluciferase and TR-FRET conditions, the living SKOV-3 cell gate targeted the PI-negative cells expressing EGFR at a bright to dim level, to take into account the previously reported EGFR downregulation in the presence of anti-HER2 antibodies.^51,52 For the calcein labeled cells, PI was not used because of its spectral overlapping with calcein, and the residual living SKOV-3 cells were gated on both the EGFR detection and the absence of calcein leakage. The data first demonstrated that the maximum lysis measured by the TR-FRET method on the culture supernatant was over-estimated, since the target cell specific lysis measured by flow cytometry was consistent with the nanoluciferase assay data. On the contrary, the flow cytometry data from the calcein-labeled cells were not significantly different from the supernatant's calcein signal measurement, reaching about 80 to 90% of specific lysis. Hence, consistently with previously published data,⁵³ our results suggest that the calcein labeling might be partially toxic for the target cells, resulting in an increased sensitivity to the Ab-induced cytotoxicity. Although these observations need to be widened in other cellular models, calcein and TR-FRET methods thus both appear inappropriate for ADCC testing in the tested models, while ⁵¹Cr- and nanoluciferase-release assays combine versatility, biological relevance and good assay performances.

Overall, the current findings demonstrate that the new nanoluciferase method is a valuable non-radioactive method that may replace the historical ⁵¹Cr-release assay for cytotoxicity measurement. It is safe, versatile, fast, easy to use and only requires simple equipment to perform. Its high sensitivity enables the use of only a very low number of target cells compared with other methods. Consequently, a limited number of effector cells is required per assay, which is particularly important, for example, in the case of patient-derived effector cells. These different advantages make the nanoluciferase method amenable to high throughput screening for limited operating costs. From a biological point of view, and in contrast to other existing non-radioactive methods, the nanoluciferase release is specific and fully correlated with target cell death. Moreover, the involved mechanisms appear to be comparable to the mechanisms involved in ⁵¹Cr release because good similarities were demonstrated between both methods in terms of the kinetics, spontaneous and maximum releases and specific lysis percentages in the ADCC and CDC models using adherent or non-adherent target cells. The full biological relevance of the nanoluciferase-mediated cell death measurement supports the regulatory recommendations and guidelines. Thus, the nanoluciferase method represents a suitable tool for the development of many cytotoxicity-related potency assays (e.g., ADCC, CDC, and apoptosis) required during the development of innovator or biosimilar therapeutic Abs, from early candidate screening to quality control testing for lot release.

Materials and methods

Cell lines and culture

All cultures were performed at 37°C in a humidified atmosphere that contained 5% CO₂. The Raji (ECACC, catalog #85011429), SKOV-3 (ATCC, #HTB-77) and CHO/TNF cell lines were cultured in RPMI-1640 (Sigma-Aldrich, #R8758), McCoy's (Sigma-Aldrich, #M8403) and Ham's F12 (Sigma-Aldrich, #N6658) media, respectively, and were all supplemented with 10% fetal bovine serum (Sigma-Aldrich, #F7524). The CHO/TNF cells corresponded to a CHO-K1 cell line (ECACC, #85051005) that stably expressed a low-cleavage mutant form of TNF, deleted from amino acids 1 to 12,⁵⁴ which was previously developed in our laboratory (unpublished). Forty-eight hours before assay performance, target cells were passaged and cultured in their specified medium.

The cytotoxic T cells genetically modified to express a CD16/FcϵRIγ chimerical receptor were used as effector cells.³⁸ These cells enabled a significant improvement in the assay precision (unpublished). Eighteen to 24 h before the ADCC assay performance, the effector cells were thawed in RPMI-1640 that contained 50% fetal bovine serum (FBS; Sigma-Aldrich, F7524), centrifuged 10 minutes at 160 g and cultured in RPMI-1640 supplemented with 10% FBS and 300 IU/mL of IL-2 (Miltenyi Biotec, 130–093–903).

Construction of nanoluciferase-expressing cell lines

Vector construction

Using the nanoluciferase sequence available in GenBank (Accession Number JQ513379), chemical synthesis was performed after codon optimization for expression in eukaryotic cells using Invitrogen's GeneArt® and GeneOptimizer™ technologies.⁵⁵ The synthesized nanoluciferase sequence was subsequently sub-cloned under the control of the human CMV Immediate Early promoter in a PiggyBac vector that contained the puromycin resistance gene (SBI, PB510B-1) to obtain the pPB-NLuc vector.

Establishment and follow-up of the nanoluciferase cell lines

The Raji, SKOV-3 and CHO/TNF cell lines were electroporated with both pPB-NLuc and PiggyBac transposase (SBI, PB210PA-1) vectors using a Neon transfection system (Invitrogen, Carlsbad, CA) and were cultured in the presence of puromycin (Invivogen, #ant-pr) at 0.25, 5 and 20 µg/mL, respectively. After 7 to 10 d of antibiotic selection, transfectants were cloned by limiting dilution, and the clones were screened based on their nanoluciferase expression following the lysis of 3,000 cells with 0.24% Triton X-100 (Sigma-Aldrich, T9284) and the resulting luminescence was read as described below.

The selected clones were maintained in culture under antibiotic selection during 50 passages, and the nanoluciferase expression was regularly monitored by measuring the nanoluciferase S/B ratio after cell lysis. Briefly, 3,000 cells were incubated for 15 minutes in a 96-well flat-bottom microplate in culture medium alone or with 0.24% Triton X-100 to determine the spontaneous and maximum release, respectively. The incubation plates were subsequently centrifuged for 1 minute at 860 g, and 25 µL of supernatant were transferred in 96 half-area white microplates (Greiner BioOne, #675074). Twenty-five µL of substrate (NanoGlo® Luciferase assay system, Promega, #N1130) diluted 1/50 in phosphate-buffered saline (PBS; Sigma-Aldrich, D8537) were added to each well; after 3 minutes of incubation at room temperature, the plates were read with a Mithras LB940 multimode microplate reader (Berthold Technologies, Bad Wildbad, Germany) with an acquisition time of 0.05 s per well. The nanoluciferase activity was expressed in RLU. The S/B ratio was further calculated as the mean of the maximum release divided by the mean of the spontaneous release. The nanoluciferase-expressing clones were the only target cells used in this study.

Apoptosis induction

The apoptosis of nanoluciferase-expressing Raji cells (50,000 cells/well) was induced using incubation with increasing doses of staurosporine (Sigma-Aldrich, #S6942) (6.4 nM to 20 µM) or UV exposure (Benchtop UV Transilluminator STX-20-M, Uvitec, Cambridge, UK) for increasing periods (10 to 60 s) in a 100 µL final volume of RPMI-1640 10% FBS. Forty-eight hours after induction, 100 µL of RPMI-1640 in the presence or absence of 0.5% Triton X-100 were added to the wells. After 15 minutes at room temperature, the plates were centrifuged, and 25 µL of supernatant per well were used for the nanoluciferase signal measurement (as described previously). The percentage of dead cells was calculated for each condition as follows: (mean RLU of the cells in medium without Triton / mean RLU of the cells in medium with Triton) x 100.

The cell pellet was resuspended in the residual volume of medium and acquired on a BD Accuri™ C6 flow cytometer (BD Biosciences, San Jose, USA) following the addition of 0.25 µM TO-PRO-3 (Molecular Probes, #T3605). Living cells were gated based on the morphologic parameters (FSC/SSC) and TO-PRO-3 negativity, using the BD Accuri C6 software.

Antibodies

All therapeutic Abs were from commercially available stocks. Anti-CD20 rituximab (MabThera®, lot number H0146) and anti-HER2 trastuzumab (Herceptin®, lot number H4406) were purchased from Roche. Anti-TNF adalimumab (Humira®, lot number 34434) was purchased from Abbott Laboratories. For the ADCC and CDC assays, antibodies were serially diluted in RPMI-1640 to obtain the indicated concentrations. For the accuracy evaluation experiment, in addition to the standard reference range (considered the “reference sample”), additional ranges that represented 80% or 125% of the standard range nominal concentrations were prepared to mimic under- and over-potent samples (considered “test samples”) for accuracy evaluation.

⁵¹Cr and nanoluciferase comparison in ADCC and CDC assays

For the ⁵¹Cr-release assay, the nanoluciferase-expressing target cells were initially incubated in the presence of 1 µCi/10⁶ cells of ⁵¹Cr (sodium chromate, PerkinElmer, NEZ030) for 1 h at 37°C and were then extensively washed before use to eliminate the excess free ⁵¹Cr. For the nanoluciferase-based assay, the nanoluciferase-expressing target cells were directly usable for the assay.

Except when otherwise stated, 3,000 target cells per well were plated in flat-bottom 96-well plates for all the ADCC and CDC assays, in a final volume of 100 µL, with each condition tested in triplicate. For the ADCC assay, the E:T ratio was 10:1, and the assay was performed in RPMI-1640 with 5% FBS. The CDC assays were performed in raw RMPI-1640 in the presence of a fixed quantity of complement per well, i.e., 1 CH₅₀ (complement hemolytic activity 50)/well for Raji cells and 6 CH₅₀/well for CHO/TNF cells, using serum from guinea pig titrated for complement activity (Sigma-Aldrich, #S1639).

For all assays, the target cells and antibodies were pre-incubated for 20 minutes at room temperature in the assay plate. Following the addition of effector cells or complement, the plates were incubated at 37°C, 5% CO₂ for the indicated time (4 h, except when otherwise stated). Each measurement plate included maximum and spontaneous release conditions, i.e., the target cells in medium containing an irrelevant Ab (rituximab for SKOV-3 cells or trastuzumab for Raji and CHO/TNF cells), effector cells or complement, and the presence or absence of 0.01% Triton X-100, respectively. At the end of the incubation period, 50 µL of supernatants were transferred in 96 V-bottom microplates and centrifuged for 1 minute at 860 x g.

For the ⁵¹Cr-labeled cells, 25 µL of supernatant were subsequently transferred in Lumaplates (PerkinElmer). The plates were dried, protected from light, for one night at room temperature before gamma counting in a MicroBeta JET plate reader (PerkinElmer, Waltham, USA). For the unlabeled target cells, the nanoluciferase signal was read as described previously.

Nanoluciferase, calcein and TR-FRET methods comparison in ADCC assays

For the calcein- and TR-FRET-based assays, nanoluciferase-expressing Raji or SKOV-3 cells were labeled with calcein-AM (Biolegend, 425201) or with the fluorescence-enhancing ligand bis-(AM)-2,2′:6′,2″-terpyridine-6,6″-dicarboxylate (BATDA, DELFIA® EuTDA cytotoxicity kit, Perkin Elmer, AD0116). Briefly, 3×10⁶ target cells were washed twice in RPMI-1640 and incubated 20 min at 37°C and at 10⁶ cells/mL with calcein-AM (2.5 µM, according to preliminary labeling optimization experiments) or with BATDA (1/400 final dilution, according to the manufacturer's instructions). Cells were then washed 3 times with RPMI 10%FBS. For the nanoluciferase-based assay, the nanoluciferase-expressing target cells were directly usable for the assay.

The 3 different target cell populations (unlabeled, calcein-labeled and BATDA-labeled) were then run in parallel in an ADCC assay as described above in the “⁵¹Cr and nanoluciferase comparison in ADCC and CDC assays” section. At the end of the incubation period (4 h), 50 µL of each supernatant were transferred in 96 V-bottom microplates and centrifuged for 1 minute at 860 x g.

For the nanoluciferase-based assay, supernatants from the unlabeled target cells were then directly measured for nanoluciferase activity as described above. For the calcein-based assay, 25 µL of supernatants from calcein-labeled cells were transferred in a black 96-well half-area plate (Greiner Bio-One, 675076) and read for fluorescence (excitation 485 nm, emission 520 nm, reading time 0.3 s) on the Mithras LB940 reader. For the TR-FRET-based method, 20 µL of supernatants from BATDA-labeled cells were transferred in the provided 96-well flat-bottom plate and 200 µL of europium solution was added in each well and incubated 15 minutes under agitation before measurement, following the manufacturer's instructions. The TR-FRET signal was read on the LB940 reader with the following parameters: excitation 320 nm, emission 615 nm, counting time 1 s (500 cycles of 2 ms including a 50 µs delay and a 400 µs reading time).

In some experiments, the SKOV-3 cell pellets were harvested to measure the remaining living target cells by flow cytometry. Triplicate wells were first treated for 4 min with 50 µL of trypsin-EDTA (Sigma-Aldrich, T3924) to detach SKOV-3 target cells. Trypsin was then inactivated by the addition of 100 µL of RPMI-1640 10% FBS and plates were centrifuged 1 min at 500 g. After supernatants removal, the cell pellets were harvested and pooled in 100 µL of PBS 2% FBS. Cells were then labeled with AlexaFluor-647 coupled anti-EGFR antibody (BD Biosciences, 563577). After 20 min of incubation at 4°C, cells were washed twice with PBS 2% FBS and directly acquired on a C6 flow cytometer. For nanoluciferase and TR-FRET conditions, cells were acquired in the presence of 0.4 µg/mL of propidium iodide (PI, eBiosciences, 00–6990–50). The percentage of living SKOV-3 cells was determined in the PI-negative/EGFR positive gate, respectively, using the BD Accuri C6 software. For calcein conditions, because of the spectral leakage of the calcein fluorescence on the PI channels, the accurate measurement of PI-positive dead target cells was not possible and the frequency of residual target cells was only measured in the calcein-positive / EGFR-positive gate.

Data analysis

The data analysis was performed similarly for the 4 read-out methods. The percentage of specific lysis for each Ab concentration was calculated according to the following formula: [(experimental signal – mean spontaneous release) / (mean maximum release – mean spontaneous release)] x 100. For flow cytometry data, the percentage of dead cells was calculated as follows: 100 – [% of living cells].

The spontaneous and maximum release values were plate-specific. The data were modeled using an unconstrained 4-parameter logistic regression, which generated 4 independent output parameters for each Ab dilution range: EC₅₀, maximal lysis (E_max), minimal lysis (E_min) and slope of the linear part. Furthermore, the percentage of spontaneous release was calculated as the ratio between the spontaneous and maximum signals x 100.

CVs, which is also referred to as the “relative standard deviation,” were calculated as [standard deviation]/[arithmetic mean].

To evaluate the performances of the methods, the EC₅₀ ratios were calculated for each 80% and 125% test sample as [EC₅₀(reference)]/[EC₅₀(test)] from the corresponding unconstrained 4-parameter logistic regression models. For the accuracy evaluation, the experimental bias (in %) was subsequently calculated as an extrapolation of the USP <1033> formula: 100 x [(Measured EC₅₀ ratio/Target EC₅₀ ratio)-1]. The target EC₅₀ ratios were 0.8 and 1.25 for the 80% and 125% test samples, respectively.

All statistical analyses, linear and nonlinear regressions and graphs were performed using GraphPad Prism® software. The specific lysis, S/B ratios and common calculations were performed using Microsoft Excel.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Acknowledgment

We are grateful to Pr. Gilles Thibault for his critical reading and helpful participation in the writing of this manuscript.

Funding

This work was performed in the frame of the “Fonds Unique Interministeriel” funded project PremiumADCC. We thank BPIfrance and Region Pays de la Loire for funding.