A whole blood in vitro cytokine release assay with aqueous monoclonal antibody presentation for the prediction of therapeutic protein induced cytokine release syndrome in humans

Cytokine 60 (2012) 828–837

Contents lists available at SciVerse ScienceDirect

Cytokine journal homepage: www.journals.elsevier.com/cytokine

A whole blood in vitro cytokine release assay with aqueous monoclonal antibody presentation for the prediction of therapeutic protein induced cytokine release syndrome in humans Babette Wolf a,1, Hannah Morgan a,1, Jennifer Krieg a, Zaahira Gani a, Adriana Milicov b, Max Warncke c, Frank Brennan a, Stewart Jones a, Jennifer Sims a, Andrea Kiessling a,⇑ a

Biologics Safety and Disposition, Preclinical Safety, Novartis Institutes for BioMedical Research, Novartis Pharma AG, Werk Klybeck, Basel, Switzerland Biologics Center, Protein Production and Antibodies, Novartis Institutes for BioMedical Research, Novartis Pharma AG, Novartis Campus, Basel, Switzerland c Department of Autoimmunity, Transplantation and Inflammation, Novartis Institutes for BioMedical Research, Novartis Campus, Basel, Switzerland b

a r t i c l e

i n f o

Article history: Received 1 May 2012 Received in revised form 16 August 2012 Accepted 18 August 2012 Available online 15 September 2012 Keywords: Cytokine release Whole blood Monoclonal antibody TGN1412

a b s t r a c t The administration of several monoclonal antibodies (mAbs) to humans has been associated with acute adverse events characterized by clinically significant release of cytokines in the blood. The limited predictive value of toxicology species in this field has triggered intensive research to establish human in vitro assays using peripheral blood mononuclear cells or blood to predict cytokine release in humans. A thorough characterization of these assays is required to understand their predictive value for hazard identification and risk assessment in an optimal manner, and to highlight potential limitations of individual assay formats. We have characterized a whole human blood cytokine release assay with only minimal dilution by the test antibodies (95% v/v blood) in aqueous presentation format, an assay which has so far received less attention in the scientific world with respect to the evaluation of its suitability to predict cytokine release in humans. This format was compared with a human PBMC assay with immobilized mAbs presentation already well-characterized by others. Cytokine secretion into plasma or cell culture supernatants after 24 h incubation with the test mAbs (anti-CD28 superagonist TGN1412-like material (TGN1412L), another anti-CD28 superagonistic mAb (ANC28.1), a T-cell depleting mAb (Orthoclone™), and a TGN1412 isotype-matched control (Tysabri™) not associated with clinically-relevant cytokine release) was detected by a multiplex assay based on electrochemiluminescent excitation. We provide proof that this whole blood assay is a suitable new method for hazard identification of safety-relevant cytokine release in the clinic based on its ability to detect the typical cytokine signatures found in humans for the tested mAbs and on a markedly lower assay background and cytokine release with the isotype-matched control mAb Tysabri™ – a clear advantage over the PBMC assay. Importantly, quantitative and qualitative differences in the relative cytokine responses to the individual mAbs, in the concentration-response relationships and the prominent cytokine signatures for individual mAbs in the two formats reflect diverging mechanisms of cytokine release and different levels of dependency on high density coating even for two anti-CD28 super-agonistic antibodies. These results clearly show that one generic approach to assessment of cytokine release using in vitro assays is not sufficient, but rather the choice of the method, i.e. applying the whole blood assay or the PBMC assay needs to be well considered depending on the target characteristics and the mechanistic features of the therapeutic mAbs being evaluated. Ó 2012 Elsevier Ltd. All rights reserved.

Abbreviations: CD, cluster of differentiation; Cs, calibrator samples; CV, coefficient of variation; Fc, fragment crystallizable; IFN, interferon; IL, interleukin; LLOQ, lower limit of quantification; mAb, monoclonal antibody; MSD, Meso Scale Discovery; PBMCs, peripheral blood mononuclear cells; PBS, phosphate-buffered saline; PEI, polyethylenimine; QCs, quality control samples; RT, room temperature; SEM, standard error of the mean; TGN1412L, TeGenero1412-like material; TNF, tumor necrosis factor; ULOQ, upper limit of quantification. ⇑ Corresponding author. Address: Novartis Pharma AG, Werk Klybeck, Klybeckstrasse 141, CH-4057 Basel, Switzerland. Tel.: +41 61 6962508. E-mail address: [email protected] (A. Kiessling). 1 These authors contributed equally to this work. 1043-4666/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.cyto.2012.08.018

1. Introduction Monoclonal antibodies and other recombinant proteins that target receptors expressed on immune cells represent promising therapeutic agents to modulate immune responses with high specificity in an increasing number of disease indications [1–4]. Consequently, there is a growing need for the development of methods for hazard identification and potentially quantitative risk assessment for on-target immunostimulatory effects in humans

B. Wolf et al. / Cytokine 60 (2012) 828–837

that can result in clinical toxicity [5]. The failure to predict the cytokine release syndrome induced by the anti-CD28 superagonist TGN1412 in humans has dramatically demonstrated the limitations of pre-clinical animal models [6] and has triggered a broad spectrum of activities to develop and qualify in vitro tests with high predictive value using human systems [7]. The test formats include assays in whole or diluted human blood and methods using isolated PBMCs with mAbs presented in aqueous or immobilized form. In order to mimic the in vivo situation for a specific mAb as closely as possible, it should be considered that whole blood in contrast to PBMCs contains certain cell populations such as erythrocytes and granulocytes as well as factors such as complement which may act as contributors to or modulators of cytokine responses. Autologous plasma containing soluble factors can be added back to the PBMC culture medium at a certain percentage, but this does not reflect their original concentrations in undiluted plasma. However, using whole blood may not always eliminate these limitations outlined for PBMC based assays, especially when interactions with endothelial cells are of importance or if the majority of target cells are located in tissues. Although PBMCs were shown to produce distinct responses upon air-dried presentation of mAbs known to induce cytokine release in vivo [8], this approach may tend to overestimate risks and produce false positive results for some antibodies such as Tysabri™ and Avastin™ that can infrequently cause infusion reactions in humans which generally lack a causative link to cytokine release syndrome. For example, Findlay et al. [8] revealed marked cytokine release with Tysabri™ using a PBMC-based assay, although Tysabri™ is associated with a relatively low incidence of infusion reactions not related to adverse cytokine release but frequently related to the development of anti-drug antibodies and hypersensitivity reactions [9,10]. In addition, the method of mAb presentation may strongly influence the results. In this context, it is questionable, whether cytokine release stimulated by high density coating, e.g. by immobilization onto the plate via air-drying, provides relevant data for risk assessment of a new mAb, when a physiological correlate to this presentation form cannot be identified. The retrospective testing of TGN1412 in different assay formats was mainly focused on reproducing the patterns and levels of cytokines seen in vivo as closely as possible. However, the finding that TGN1412 must be presented in an immobilized form for hazard identification [11] raises the question as to what the in vivo correlate could be. In an assay using PBMCs or diluted blood (20% v/v) with TGN1412 added to the aqueous phase, very small responses almost similar to the isotype control were found [11,12]. There is the possibility that responses were not detected upon aqueous presentation due to the high dilution of human blood. The aim of our study was to evaluate a cytokine release assay in almost undiluted whole blood in combination with aqueous mAb presentation for hazard prediction and potential risk assessment of in vivo cytokine release by mAbs with respect to cytokine pattern, magnitude, frequency and concentration dependency of responses. To demonstrate the value of this assay as part of a new in vitro cytokine release testing strategy, we conducted a comparative study with a PBMC assay previously shown to predict TGN1412-induced cytokine release upon air-dried coating following the assay protocol published by Findlay et al. [12] as close as the published method description allowed. However, the quantitative determination of secreted cytokines – a commercial multiplexed plate based assay coupled to electrochemiluminiscence detection – differed from the described method, since the authors used ELISA methods with in-house generated recombinant cytokine standards and raised anti-cytokine antibodies not accessible to us. We evaluated internally produced TGN1412-like material (sequence-identical material; TGN1412L), another super-agonistic mouse anti-human CD28 mAb, and Orthoclone™ – a murine

829

anti-human CD3 mAb whose mechanism of cytokine release resembles that of TGN1412 in that the target-bearing cells are expected to be the main source of cytokine release [13]. Tysabri™ was used as an isotype-matched control for TGN1412L in all assays. 2. Materials and methods 2.1. Generation of vector constructs for TGN1412 heavy and light chain Human heavy and light chain constant regions and TGN1412 heavy and light chain variable regions were synthesized by Geneart (accession numbers of human IgG4: P01861; human kappa light chain: P01834). Human heavy and light chain constant regions were cloned directly after the anti-CD28 TGN1412 variable region to generate the vector constructs [pRS5a-TGN1412-human IgG4 heavy chain] and [pRS5a-TGN1412-human kappa light chain], respectively. 2.2. Transient expression of anti-CD28-TGN1412-hIgG4/kappa HEK293 Freestyle cells (Invitrogen, Carlsbad, CA, USA) were cultivated in Freestyle 293 medium (Invitrogen) in Roller bottles at 6.5–7.5 rpm, 5% CO2 and 37 °C to a density of 3.17  106 viable cells/ml. The cells were transiently transfected with a vector DNA:PEI (Polyscience, Warrington, PA, USA) mixture, using a 1:3 ratio. For this, [pRS5a-TGN1412-human IgG4 heavy chain], [pRS5a-TGN1412-human kappa light chain] and PEI were diluted in M11V3 medium (Novartis, Basel, Switzerland), sterile-filtered and combined. After an incubation for 10 min at RT, the DNA:PEI mixture was added to the cells. Six hours after transfection, Freestyle 293 medium was added to the culture. The cells were then further cultivated at 6.5– 7.5 rpm, 5% CO2 and 37 °C. Eleven days after transfection, cells were harvested and the supernatant was sterile filtered through a stericup filter (0.22 lm; Stericup Express™ Plus, Millipore, Billerica, MA, USA). The sterile supernatant was stored at 4 °C. 2.3. Purification of anti-CD28-TGN1412-hIgG4/kappa Purification was performed on an ÄKTA 100 explorer air chromatography system (GE Healthcare, Glattbrugg, Switzerland) at 4 °C in a cooling cabinet, using a freshly sanitized (0.25 M NaOH) HiTrap ProtA MabSelectSuRe column (GE Healthcare). The column was equilibrated with Dulbecco’s PBS, and then the sterile-filtered supernatant was loaded at 4.0 ml/min. After washing with DPBS, the anti-CD28 mAb was eluted with 50 mM Citrate, 70 mM NaCl pH3.2. The eluate was collected in fractions, pooled, adjusted to pH6 with 1 M Tris HCl pH10 and sterile filtered (0.22 lm; Millipore Steriflip). For the second purification step, pools from the first purification were loaded into a freshly sanitized (0.5 M NaOH) and equilibrated (50 mM Citrate, 140 mM NaCl pH6) Sephadex column (Hi Load 16/60 Superdex 200 grade) (GE Healthcare), and the run was performed with the same buffer at 1 ml/min. The total amount was divided in two different pools, one in 50 mM Citrate/140 mM NaCl pH6 and the second pool was neutralized at pH7 with 1 M Tris HCl pH10. These two different formulations were tested in the cytokine release assays. The optical density at 280 nm of the pooled eluate was measured in a spectrophotometer ND-1000 (NanoDrop), and the protein concentration was calculated based on the sequence data. The final material was tested for purity by SDS-polyacrylamide gel electrophoresis which shows purity above 90%. Limulus amebocyte lysate test revealed an endotoxin level below 0.25 EU/mg. The mAb sequence was verified by mass spectromety. Aggregation

830

B. Wolf et al. / Cytokine 60 (2012) 828–837

was analyzed by size exclusion chromatography coupled with multi-angle light scattering detector (SEC-MALS), as described in Section 2.4. Finally, binding to CD28 on lymphocytes was confirmed by FACS analysis. 2.4. Size exclusion chromatography coupled with multi-angle light scattering detector (SEC-MALS) SEC-MALS measurements were performed on an Agilent 1200 HPLC system (Agilent Technologies) connected to a tri-angle light scattering detector (miniDAWN Treos, Wyatt Technology, Santa Barbara, CA, USA). The concentration of the sample was followed online with a differential refractometer (Optilab rEX, Wyatt Technology) using a specific refractive index increment (dn/dc) value of 0.186 ml/g. Sample volumes of 50 ll were injected on a Superdex 200 10/300 GL column (GE Healthcare). The data were recorded and processed using the ASTRA V software (Wyatt Technology). To determine the detector delay volumes and normalization coefficients for the MALS detector, a BSA sample was used as reference. The obtained results were 0.8% aggregates, 98.4% monomers and 0.9% degradation. 2.5. Monoclonal antibodies Two formulations of the anti-CD28 TGN1412L were produced in-house as described above. The clinical grade anti-CD3 mAb Orthoclone™ (murine IgG2a, Muromonab; Janssen Cilag, Baar, Switzerland) was used as a positive control as it has been associated with clinically relevant cytokine release syndrome. As a second positive control, ANC28.1 (murine IgG1 super-agonistic anti-CD28 mAb; Calbiochem, Merck, Darmstadt, Germany) was included. Tysabri™, a clinical grade human monoclonal mAb directed against a4-integrin (Natalizumab; Biogen Idec, Zug, Switzerland) of the same isotype (IgG4) as TGN1412 was used as an isotypematched control. 2.6. Blood donors Human blood (in lithium heparin vacutainersÒ; BD, Allschwil, Switzerland) was taken by consent from healthy donors in accordance with local ethical practice. Blood was stored at room temperature and used within 2 h after donation. 2.7. Cytokine release assays in whole blood Monoclonal antibodies were serially diluted in sterile PBS (pH7.4; Gibco, Life Technologies, Zug, Switzerland) and added at a volume of 12.5 lL in triplicate to the wells of U-bottomed polystyrene microtiter plates (Corning, Kaiserslautern, Germany). Then, 237.5 lL of whole blood were added to reach the indicated final concentrations (lg/ml) while avoiding additional pipetting for mixing which could lead to cell activation. The selected concentrations were judged optimal based both on published literature and internal experience. Plates were incubated for 6, 24 or 48 h (time course experiments) or 24 h (selected incubation time based on results of time course experiments) at 37 °C, 5% CO2. Following incubation, plates were centrifuged at 1800 g and plasma supernatants were collected and transferred to fresh microtiter plates. Plasma supernatants were freshly analyzed or stored for up to one week at 80 °C before cytokine determination. 2.8. Isolation of PBMCs PBMCs were isolated by density gradient centrifugation using Biocoll (Biochrom AG, Berlin, Germany) and Leucosep tubes (Greiner Bio One, Frickenhausen, Germany). Whole blood was diluted 1:4

with sterile PBS (pH7.4; Gibco) and centrifuged at 980 g without brake. The isolated PBMCs were washed 2–3 times with cold PBS at 360 g. After washing, the cells were counted and resuspended in pre-warmed RPMI1640 cell culture medium (Gibco) substituted with 1% L-glutamate (Gibco) and 2% donor plasma at a concentration of 5  105 cells/ml. 2.9. Cytokine release assay using PBMCs MAb dilutions prepared in sterile PBS (pH7.4; Gibco) were added at 50 ll/well and air-dried onto the surface of U-bottomed polystyrene microtiter plates by overnight incubation at RT resulting in the indicated total amount (lg/well). Prior to the addition of isolated PBMCs, microtiter plates were washed with sterile PBS. 1.25  105 cells per well (250 lL) were added in triplicate to the air-dried plates and incubated for 24 h at 37 °C, 5% CO2. Following incubation, plates were centrifuged at 360 g at RT and cell culture supernatants were collected and transferred to fresh microtiter plates. Plasma supernatants were freshly analyzed or stored for up to one week at 80 °C before cytokine determination. These conditions for storage were internally evaluated before this study. 2.10. Cytokine detection IFN-c, TNF-a, IL-2, IL-4, IL-6, IL-8 and IL-10 in either the plasma (for whole blood) or cell culture supernatants (for PBMCs) were detected using a custom human 7-plex assay kit from MSD (Meso Scale Discovery, Gaithersberg, MD, USA). All steps were performed at RT and with shaking at 700 rpm (Titramax 1000, Heidolph, Schwabach, Germany). Plates were incubated with MSD kit diluent 2 for 30 min before the addition of unknown samples, Cs, QCs and blank samples and then incubated for a further 2 h. Triplicate unknown samples were analyzed as derived from the cytokine release assay together with duplicate determinations of Cs, QCs and blank samples. Cs and QCs solutions were diluted in MSD kit diluent 2 for plates measuring plasma supernatants or cell culture medium for plates measuring PBMC cell culture samples. Following sample incubation, the plates were washed four times with washing buffer (PBS/0.05% Tween-20). A cytokine-specific detection antibody mix (concentration for each antibody of 1 lg/ml except for anti-IL-8 used at 0.1 lg/ml) in MSD kit diluent 3 was added to the plates and incubated for 1 h. Plates were washed as previously described and 2x read buffer (MSD) was added. Measurement of electrochemiluminescence was performed on the Sector 2400 instrument (MSD) and the cytokine concentrations were determined by back calculation to the calibrator standard curve fitted by a 4PL approach. Cs were prepared in a concentration range of 5000–1.2 pg/ml. Two QCs at the concentrations 1250 and 19.5 pg/ml were included on each plate. The CVs on duplicate concentrations for Cs and QCs were 620%. The accuracy of mean duplicate concentrations was 80–120% and 70–130% for Cs and QCs, respectively. The LLOQ and ULOQ of the assay were determined as 9.8 pg/ml and 5000 pg/ml during assay validation. 3. Results 3.1. Aqueous presentation of TGN1412L to whole blood 3.1.1. Time course analysis In order to determine the most appropriate time point for cytokine measurements, a kinetic study was conducted using the whole blood cytokine release assay. For this, whole blood from three healthy human donors was stimulated in vitro with two different formulations of TGN1412L (pH6 and pH7) at 0.1, 1.0, 10 and 100 lg/ml in aqueous presentation form. The two different

B. Wolf et al. / Cytokine 60 (2012) 828–837

formulations of TGN1412L were included to assess the impact of pH on the assays. Plasma supernatants were collected after 6, 24 and 48 h for measurement of cytokines associated with cytokine release syndrome in vivo (IFN-c, TNF-a, IL-2, IL-4, IL-6, IL-8 and IL-10). The results for a donor strongly responding to TGN1412L are shown in Fig. 1. As isotype and negative controls, equal concentrations of isotype-matched mAb Tysabri™ and PBS were included. In addition, a super-agonistic anti-CD28 mAb (ANC28.1) at 5, 0.5 and 0.05 lg/ml and an agonistic anti-CD3 mAb (Orthoclone™) that primarily causes target cell-mediated cytokine release were

831

analyzed at 0.5 and 0.05 lg/ml for the same time points. Stimulation for 6 h already showed a marked IL-2 release by TGN1412L with both formulations with no further accumulation at 24 h and 48 h at concentrations P10 lg/ml. This result is in full accordance with the mechanistic action of TGN1412 on target T-cell populations and has been linked to the induction of CD4 + T-cell proliferation by TGN1412 as previously demonstrated in assay formats using immobilized TGN1412 in combination with PBMCs [11]. All other TGN1412L-stimulated cytokines were low at 6 h and peaked or reached plateau at 24 h at concentrations of 10 and 100 lg/ml

Fig. 1. Time course analysis of cytokine release in a whole blood assay with aqueous mAb presentation. (A–G) Cytokine responses from whole blood plasma supernatants after 6 (clear bars), 24 (diagonal lined bars) or 48 h (filled bars) stimulation with the monoclonal antibodies ANC28.1, Orthoclone™, Tysabri™ and the two formulations of TGN1412L. PBS served as a negative control. Stimulator concentrations are given as lg/ml. The figure illustrates data from one strongly responding donor.

832

B. Wolf et al. / Cytokine 60 (2012) 828–837

except for IL-6 which still accumulated up to 48 h. For most cytokines and tested concentrations, the ANC28.1- and Orthoclone™induced cytokines accumulated frequently up to 48 h in relation to the PBS negative control and Tysabri™, demonstrating differences in kinetics for these antibodies. However, 48 h incubation revealed remarkably elevated concentrations of several cytokines, especially of IL-8, in PBS and the isotype control. Therefore, the 24 h stimulation period was deemed optimal for subsequent screening experiments. 3.1.2. TGN1412L-induced cytokine pattern after 24 h The cytokine release from the whole blood of 20 healthy human donors was determined as described in Section 3.1.1 for the selected incubation time of 24 h, and the resulting means and SEM values are summarized for all cytokines in Fig. 2A–G. Between the strongest and weakest responding donors a wide range of detectable cytokine concentrations was found indicating a high inter-donor variability for all cytokines as depicted in Supplementary Fig. 1 and Table 1. Tested concentrations of TGN1412L (both formulations) above 0.1 lg/ml caused a markedly higher response compared with the respective concentration-matched Tysabri™ control for all tested cytokines in responsive donors. In general, incubation with PBS revealed no concentration levels above the LLOQ for any cytokine with the exception of IL-8 which consistently gave high background responses. However, the IL-8 background levels were still below the overall responses to TGN1412L (both formulations) and the positive control antibodies except for TGN1412L pH6 at 0.1 lg/ml and ANC28.1 at 0.05 lg/ml. On average the highest tested concentration of ANC28.1 (5 lg/ml) induced much higher levels of any cytokine than all tested TGN1412L concentrations, while at lower concentrations of ANC28.1 (0.5 and 0.05 lg/ml) cytokine responses were within the same range as seen with TGN1412L. Overall cytokine responses to Orthoclone™ at 0.5 and 0.05 lg/ml were higher than measured after stimulation with TGN1412L at all tested concentrations for the majority of donors with maximum levels at 0.05 lg/ml except for IL-4 and IL-8. Overall release of IL-2, a typical signature cytokine for Orthoclone™, was about three times higher at 0.05 lg/ml when compared to 0.5 lg/ml potentially indicating a more pronounced T cell stimulatory capacity at the lower concentration as opposed to greater activation-based apoptosis and anergy at higher concentrations in this assay setup. In order to illustrate the cytokine pattern and overall donor responsiveness induced by TGN1412L when presented in aqueous form in whole blood, the frequency of donors responding above the isotype control Tysabri™ at matched concentrations is depicted in Fig. 2H. More than half of the tested donors showed specific release of IL-2 and IL-6 in response to TGN1412L, 50% of donors responded specifically with IL-10 release, whereas TNF-a, IFN-c and IL-4 were secreted by only 20–25% of screened donors. All donors showed a response above Tysabri™ for IL-8. 3.2. Solid phase presentation of TGN1412L to PBMCs Fig. 3A–G shows means and SEM of different cytokines released into cell culture supernatants after 24 h stimulation of PBMCs derived from ten healthy human donors with immobilized TGN1412L – an assay format already well-characterized and applied by others [11,12]. Evaluations were performed in direct comparison with the corresponding positive and negative controls as done for the whole blood assay, where the final concentrations (lg/ml) for aqueous presentation in the whole blood assay were matched with the total immobilized mAb amounts in the PBMC assay (lg/well). In accordance with the whole blood assay and published literature, a 24 h incubation was considered optimal. Concentration dependent cytokine release could be observed for all cytokines reaching their

highest values at 100 lg/well air-dried TGN1412L and Tysabri™ for most donors. As seen in whole blood, the two different TGN1412L formulations resulted in comparable cytokine patterns and levels with a high variability amongst individual donor responses indicating no influence of the pH 6-7 TGN1412L formulations on the cytokine responses (Supplementary Table 2). Overall cytokine responses to all stimulators were clearly higher in supernatants from PBMCs when compared to plasma supernatants from whole blood. Stimulation with the isotype control mAb Tysabri™ induced elevated levels of TNF-a, IL-6 and IL-8 above the LLOQ (most prominent at 100 lg/well) and increased responses for IL-2 (especially at 0.1 lg/well) in P50% of the donors when compared to the background signals in PBS. Moreover, background responses in the absence of a coated mAb (PBS control) were frequently observed above the LLOQ for TNF-a, IL-6 and IL-8. As a consequence of the increased assay background reflected in the observed responses to Tysabri™ at 0.1 lg/well, the responses to TGN1412L at the lowest tested concentration of 0.1 lg/well were in general not significantly higher than the concentrationmatched Tysabri™ control (paired, two-tailed Student‘s t-test, p < 0.05). As for TGN1412L, a clear concentration dependency could be observed for ANC28.1 in most donors with the highest release observed at 5 lg/well – the highest tested concentration. Notably, PBMCs stimulated by air-dried TGN1412L, ANC28.1 and Orthoclone™ evoked only a weak response (below 25 pg/ml) for IL-4 in two donors for TGN1412L and ANC28.1 and in four donors for Orthoclone™. Stimulation of PBMCs with 5 lg/well of the superagonist ANC28.1 revealed on average similar cytokine levels as detected for TGN1412L at 100 lg/well, especially for TNF-a, IL-6 and IL-8. This finding is in contrast to the cytokine release measured after aqueous ANC28.1 incubation at 5 lg/ml in whole blood which clearly exceeded the TGN1412L response at all tested concentrations. Immobilized presentation of Orthoclone™ to PBMCs at 0.5 lg/well resulted in comparable average cytokine concentration levels for TNF-a, IL-2, IL-4, IL-6 and IL-8 as for TGN1412L at 100 lg/well in contrast to the cytokine pattern seen in whole blood where TGN1412L responses were generally lower than with Orthoclone™. Moreover, Orthoclone™-induced cytokine levels were similar or lower at 0.05 lg/well when compared to 0.5 lg/well in the PBMC assay indicating mechanistic differences between the two assay formats. Remarkably, in contrast to the whole blood assay overall release of IL-2 was more than two times higher at 0.5 lg/well when compared to 0.05 lg/well, possibly indicating that enhanced cross-linking by immobilization at higher concentrations counteracts apoptosis as previously documented by Carpenter et al. [14], who showed that cross-linking of non-Fcgamma receptor-binding Orthoclone™ variants by antiglobulin enhanced T-cell receptor internalization and minimized induction of T-cell apoptosis. The number of responding donors summarized in Fig. 3H is in general higher for all cytokines compared to the whole blood assay. TNF-a, IL-2, IL-6 and IL-10 could be specifically detected above the concentration-matched Tysabri™ control in all donors after TGN1412L stimulation. 70% and 50% of screened donors showed specific responses to TGN1412L on IFN-c and IL-10 levels, respectively.

4. Discussion 4.1. Clinical relevance of the whole blood cytokine release assay Recent approaches to develop in vitro cytokine release assays that best identify the hazard for cytokine release syndrome in

B. Wolf et al. / Cytokine 60 (2012) 828–837

833

Fig. 2. Whole blood cytokine release determination in 20 human donors. (A–G) Cytokine responses from whole blood plasma supernatants after 24 h stimulation with the monoclonal antibodies ANC28.1, Orthoclone™, Tysabri™ and the two formulations of TGN1412L. PBS served as a negative control. Stimulator concentrations are given as lg/ ml. Values are mean ± SEM of responses from 20 donors. (H) Number of donors with cytokine responses to TGN1412L above PBS and the matched concentration of Tysabri™.  P < 0.05, P < 0.01; statistics generated by paired, two-tailed Student‘s t-test between responses to TGN1412L and the responses to the matched concentration of Tysabri™.

humans are based on the goal to optimally replicate the cytokine release patterns, levels and release kinetics for mAbs already known to be associated with cytokine release syndrome in clinical trials. These include Orthoclone™, Campath and TGN1412, the latter being the most intensively studied so far [8,11,12,15,16]. The design of the assays for such retrospective evaluations of wellknown inducers of cytokine release largely benefits from the

increasing body of understanding regarding the involved cell populations and mechanisms of cytokine release. A much larger challenge, of course, is the correct prospective hazard identification for mAbs that have not yet undergone clinical testing and where there may be concern for cytokine release. With respect to clinically relevant testing, the goal is to identify potentially hazardous mAbs associated with in vivo cytokine release associated with adverse

834

B. Wolf et al. / Cytokine 60 (2012) 828–837

Fig. 3. Cytokine release assay with PBMCs and air-dried mAb presentation in 10 human donors. (A–G) Cytokine responses from PBMC cell culture supernatants after 24 h stimulation with the monoclonal antibodies ANC28.1, Orthoclone™, Tysabri™ and the two formulations of TGN1412L material. PBS served as a negative control. Stimulator concentrations are given as lg/well. Values are mean ± SEM of responses from 10 donors. (H) Number of donors with cytokine responses to TGN1412L above PBS and the matched concentration of Tysabri™. P < 0.05, P < 0.01; statistics generated by paired, two-tailed Student‘s t-test between responses to TGN1412L material and responses to the matched concentration of Tysabri™.

events prior to initiating clinical testing and not those which may cause mild infusion reactions with different causalities. In this context, in vivo-relevant cytokine release data may also have an important contribution to the safe starting dose selection by informing the Minimum Anticipated Biological Effect Level (MABEL), and dose escalation in first-in-human trials [17]. In this study, the testing approach in almost undiluted human blood combined with aqueous mAb presentation was shown to

represent a useful and promising assay with only minimal sample and mAb manipulation aiming to preserve the naturally occurring conditions in vivo. In general, the value of in vitro and ex vivo tests in whole blood – although mostly described for diluted blood so far – has already been demonstrated for pyrogenicity testing as well as potency testing of immunostimulants and immunosuppressants with good correlation to in vivo data and reproducibility of results upon assay transfer to other laboratories [18]. Such studies have

B. Wolf et al. / Cytokine 60 (2012) 828–837

already underpinned the predictive value and reliability of simple and economic whole blood tests in other areas and are in good alignment with the results of our study. 4.2. The use of the whole blood cytokine release assay in the context of other testing approaches for prediction of cytokine release in humans Several approaches of in vitro assay testing have achieved distinct cytokine responses to model mAbs, but these frequently use relatively artificial presentation conditions including air-dried coating of antibodies, anti-Fc capture-based presentation of antibodies [11] or high density mAb coupling to beads [19]. Although cross-linking and high density presentation of mAbs simulated by these methods may have a physiological correlate in vivo in specific cases, e.g. clustering of the target via the interaction of the Fc part of antibodies with Fc gamma receptors on other immune cells or the cross-linking of mAbs by anti-drug antibodies, there is doubt about the physiological relevance in other cases. For TGN1412, the described absolute dependency of cytokine release induction on surface immobilization of the mAb still lacks a confirmed in vivo explanation, although some speculation about the potential mechanisms has been made [8]. Consequently, there has been some concern that these approaches may overestimate the risk of new mAbs under investigation by the generation of false-positive results [13]. One reason for the failure to induce cytokine release by aqueous TGN1412 presentation was possibly the use of PBMCs or highly diluted blood, not almost undiluted whole blood that may resemble the in vivo conditions better. Another potential predictive gap may be associated with the use of PBMCs as a cellular source, since these purified cells lack populations present in vivo and may, moreover, contribute to unspecific responses and increased background due to pre-activation. The use of whole blood may be a prerequisite here, although both mentioned cellular sources may lack the specific target cell populations that mediate cytokine release, when these are localized in tissues or only present at very low frequencies in blood. Moreover, many cytokine release testing formats suffer from the difficulty to reproduce test results across labs, which may indicate a lack of robustness and therefore result in difficulties for inter-laboratory assay transfer. For example, dried coating of mAb material to the bottom of plates might affect structural features and expose epitopes not present in clinical situations, additionally the necessary washing procedure to remove salts from the wells before the addition of the cells may lead to variability in the amount of coated mAb. 4.3. Characterization of the whole blood assay in a comparative study with a well-described PBMC assay utilizing air-dried mAb coating A central purpose of our study was to evaluate whether the whole blood assay only minimally diluted by the mAbs in aqueous presentation, which requires only minimal manual manipulation, would be able to identify the hazard for TGN1412-mediated cytokine release syndrome and for mechanistically related antibodies, even if it does not replicate the cytokine levels detected in vivo, but is able to produce the typical signature cytokines. This understanding is in line with the common view that these assays should not be used for risk quantification, but pure hazard identification, as it is currently not possible to define a threshold or limit for cytokine release above which concern would be raised [7]. To describe the characteristics of this assay, especially with respect to cytokine patterns, occurrence of false-positive results, general assay background and the relationship of cytokine levels between the different tested antibodies, a previously described and wellcharacterized PBMC assay in combination with air-dried coating was used for comparison [12].

835

Since the original TGN1412 material was not available to our group, TGN1412L material based on the published sequence was produced and extensively characterized for purity, aggregation, degradation products and endotoxin content as specified in Sections 2.3 and 2.4. In addition, binding to CD28 on target cells in blood was confirmed by flow cytometry (data not shown). TGN1412L in two tested pH formulations induced the typical signature cytokines of TGN1412 in both cytokine release assay formats [11]. Especially IL-2, an important mediator of the acute clinical symptoms in vivo [6], was frequently already detected after 6 h at levels similar to the concentrations measured after 24 h incubation in the whole blood assay. In contrast to the published findings [11,12] that TGN1412 requires immobilized presentation to induce cytokine release, we were able to show cytokine release for IL-2, IL-6 and IL-8 in the majority of tested donors in the whole blood assay in combination with aqueous presentation. Although the cytokine levels detected were generally lower than observed in the PBMC assay with immobilized presentation, the hazard would also have been qualitatively detected using our whole blood assay format. Interestingly, a recent publication describes cytokine release induction by TGN1412 in the aqueous phase as well, but in combination with PBMCs instead of whole blood and a tissue-like pre-incubation of PBMCs at a high cell density [16]. Moreover, aqueous presentation of TGN1412 in a three-dimensional biomimetic vascular test bed containing endothelial cells grown on a collagen scaffold, whole leukocytes and 100% autologous plasma induced typical signature cytokines, indicating that the presence of all relevant cell populations and soluble factors may, in accordance with our results, overcome the dependency on high density coating [20]. Both assays evaluated in our study show a great variability in the individual donor responses for all tested antibodies inducing cytokine release in vitro, demonstrating the importance of testing a significant number of donors (P20) for hazard identification of a new mAb to characterize the range of responses accordingly. The PBMC assay clearly showed increased background responses without mAb incubation when compared with the whole blood assay, especially for TNF-a, IL-6 and IL-8. This feature may complicate the detection of lower mAb-specific responses with potential in vivo relevance as indicated by the results for TGN1412L at a concentration of 0.1 lg/well that were indistinguishable from background levels in the overall analysis for most cytokines. The control background responses observed in our PBMC assay were similar or lower than the control responses detected by Findlay et al. [8], a reference paper that allows a direct comparison of data. Moreover, the isotype-matched control mAb Tysabri™ at 100 lg/well resulted in cytokine release above the background for several cytokines in many donors, whereas responses to Tysabri™ above the LLOQ were barely detected in the whole blood assay. These results may indicate a lower false-positive rate of the whole blood assay and are consistent with negative results obtained for other mAbs that do not cause cytokine release-mediated infusion reactions in vivo according to current understanding (unpublished internal data). The data published and tabulated in detail by Findlay et al. [8] for the PBMC assay in the context of air-dried mAb coating allow a comparison of the ratios of mean absolute cytokine responses obtained in the presence of 10 lg/well TGN1412™ and isotype control Tysabri™ tested at the same concentration in our PBMC assay for IL-2, TNF-a, IL-6 and IL-8. This relative evaluation is meaningful, since it compensates for different responses to the isotype control in individual donors. The results of TGN1412L™ (pH6 selected)/Tysabri™ ratios at 10 lg/well are 266, 4.0, 2.8 and 1.9, and 270, 17.4, 11.9 and 3.5 for IL-2, TNF-a, IL-6 and IL-8, in our study and in the study published by Findlay et al. [8], respectively. Although the cytokine pattern is similar, the absolute ratios markedly differ except for IL-2 which is seen

836

B. Wolf et al. / Cytokine 60 (2012) 828–837

as the key mechanistic cytokine for TGN1412 action. These differences may be due to factors such as the biological variability between the donors and/or methodological differences. The latter might be small handling differences in the cytokine release assay itself as well as differences in the detection systems used for quantitative cytokine determination. In an internal comparative study between MSD and a Luminex-based method both developed for human cytokines, marked differences in the results were observed for most cytokines in one and the same sample (unpublished data). A major impact here may be how comparable the binding of recombinant standards included in the assay kits is to the binding of human cytokines in the samples. Moreover, based on the results published by Findlay et al. [8] for the PBMC assay in the context of aqueous mAb presentation, a comparative evaluation of the use of PBMCs versus whole blood can be made. The ratios of mean absolute cytokine responses obtained in the presence of 10 lg/ml TGN1412™ (pH6 material results of our study used as representative example) and of the concentration-matched isotype control Tysabri™ were 11.4 (IL2), 4.6 (TNF-a), 10.7 (IL-6) and 6.4 (IL-8) for the whole blood assay and 1.5 (IL-2), 1.3 (TNF-a), 4.1 (IL-6) and 0.8 (IL-8) for the above mentioned PBMC assay [8]. This indicates a clear prerequisite of whole blood when compared to PBMCs for aqueous presentation. The estimated plasma concentration of 1 lg/ml reached at the initial dose of 0.1 mg/kg of TGN1412 in the London clinical trial is known to saturate up to 80% of the CD28 target and mediate the maximum biological response [21]. In alignment with this observation, the in vitro cytokine responses in the whole blood assay were low at 0.1 lg/ml of TGN1412L, markedly increased at 1 lg/ml, and did not further increase at concentrations higher than 1 lg/ml. In contrast, the PBMC assay showed a different concentration-dependency with highest cytokine levels detected after incubation with at 100 lg/well of TGN1412, possibly due to the additional cross-linking effect caused by high density coating. The comparison of cytokine responses to the positive control mAbs with TGN1412L revealed marked differences in the detected levels in both assays. As an example, ANC28.1 at a concentration of 5 lg/ml resulted in a much higher overall release of all cytokines when compared with TGN1412L at all tested concentrations (up to 100 lg/ml) in the whole blood assay, despite both mAbs acting as anti-CD28 superagonists [22]. In contrast, the PBMC assay did not show this strong tendency with similar overall TNF-a, IL-6 and IL-8 responses to TGN1412L at 100 lg/ml when compared to ANC28.1 at a concentration of 5 lg/ml. Although ANC28.1 was a more potent inducer of cytokine release than TGN1412L in both assays, the differences obtained in both assays may result from a specific mechanistic divergence related to the respective targeted epitopes that might differentially affect the impact of intensive cross-linking on cytokine release. This example shows that mechanistic differences of closely related mAbs may affect the in vitro assay outcome in different assay formats. Moreover, divergent dose-response relationships in cytokine release for a mAb in two in vitro assay formats as seen for Orthoclone™ at two different concentrations may contribute to mechanistic conclusions as specified in Section 3.2. These examples underpin the relevance of including more than one assay format for the assessment of a novel mAb, as long as each format used has potential correlating scenarios in vivo. As an example, depending on the target biology, disease indication and the backbone structure of a mAb, intensive clustering and cross-linking via Fc receptor bearing cells or anti-drug antibodies can be an in vivo relevant scenario justifying the assessment of cytokine release by high density mAb coating in addition to soluble mAb presentation. In addition, T-cell subpopulations may be differentially stimulated in specific assay formats as shown by higher overall levels of IL-4 and IL-10 released in response to 5 lg/ml ANC28.1 in the

whole blood assay, indicating a more pronounced stimulation of type 2 helper T cells and regulatory T cells. Therefore, the use of more than one cytokine release assay format in combination with the results of other tests such as the assessment of agonistic action, apoptosis and anergy induction, target internalization and signaling in different leukocyte subsets as well as mAb-dependent and complement-dependent cellular toxicity may help to form a more complete mechanistic understanding of a novel mAb with the potential for classification into higher or lower risk profiles prior to commencing clinical trials. 4.4. Summary and future considerations In summary, we have characterized a new, only minimally undiluted whole blood cytokine release assay in combination with aqueous presentation using TGN1412L, ANC28.1, Orthoclone™ and Tysabri™ as model mAbs in comparison with a PBMC assay with immobilized mAb presentation. The whole blood assay was found suitable for hazard identification for the tested mAbs in this study, and has also previously been shown to detect cytokine release by Campath both by our group (unpublished observations) and others [23]. The whole blood assay revealed lower background responses and cytokine release to the TGN1412L-matched isotype control mAb Tysabri™ in comparison to the PBMC assay, but the donor response rates were lower in the whole blood assay. Importantly, the different relative response levels to the individual antibodies, concentration-response relationships and prominent cytokine patterns for individual therapeutic mAbs in the two formats reflect diverging mechanisms of cytokine release and different levels of dependency on high density coating and cross-linking even for two anti-CD28 superagonistic antibodies. These results clearly indicate the importance to simulate the in vivo relevant mechanistic features when selecting certain in vitro assay formats for the prediction of cytokine release of a novel mAb and the importance of integrating all available data of target biology and mechanistic studies of mAbs in the assay selection. The differences observed in the results of our PBMC assay in combination with dry-coating compared to published data on the same assay may indicate suboptimal reproducibility of the method in different labs due to a lack of assay robustness. This may be a prompt to focus more attention on the development and validation of more easily transferrable formats with a reduced number of manipulation steps to allow correct comparisons of different cytokine release outcomes of various biological drugs between different studies and labs improving the hazard and risk evaluation. The whole blood assay with aqueous mAb presentation and a minimum level of manipulation may have this capability which needs to be evaluated by inter-lab comparisons in the future. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.cyto.2012.08.018. References [1] Genovese MC. Biologic therapies in clinical development for the treatment of rheumatoid arthritis. J Clin Rheumatol 2005;11:45–54. [2] Menter A. The status of biologic therapies in the treatment of moderate to severe psoriasis. Cutis 2009;84:14–24. [3] Rommer PS, Strüve O, Goertsches R, Mix E, Zettl UK. Monoclonal antibodies in the therapy of multiple sclerosis: an overview. J Neurol 2008;255:28–35. [4] Melero I, Hervas-Stubbs S, Glennie M, Pardoll DM, Chen L. Immunostimmulatory monoclonal antibodies for cancer therapy. Nat Rev Cancer 2007;7:95–101. [5] Brennan FR, Morton LD, Spindeldreher S, Kiessling A, Allenspach R, Hey A, et al. Safety and immunotoxicity assessment of immunomodulatory monoclonal antibodies. MAbs 2010;2:233–55.

B. Wolf et al. / Cytokine 60 (2012) 828–837 [6] Suntharalingam G, Perry MR, Ward S, Brett SJ, Castello-Cortes A, Brunner MD, et al. Cytokine storm in a phase 1 trial of the anti-CD28 monoclonal mAb TGN1412. New Engl J Med 2006;355:1018–28. [7] Vidal JM, Kawabata TT, Thorpe R, Silva-Lima B, Cederbrant K, Poole S, et al. In vitro cytokine release assays for predicting cytokine release syndrome: the current state-of-the-science. Report of a European medicines agency workshop. Cytokine 2010;51:213–5. [8] Findlay L, Eastwood D, Ball C, Robinson CJ, Bird C, Wadhwa M, et al. Comparison of novel methods for predicting the risk of pro-inflammatory clinical infusion reactions during monoclonal antibody therapy. J Immunol Methods 2011;371:134–42. [9] Hellwig K, Schimrigk S, Fischer M, Haghikia A, Müller T, Chan A, et al. Allergic and nonallergic delayed infusion reactions during natalizumab therapy. Arch Neurol 2008;65:656–8. [10] Camacho-Halili M, George R, Gottesman M, Davis-Lorton M. An approach to natalizumab hypersensitivity: a case series of induction of tolerance. Mult Scler 2011;17:250–3. [11] Stebbings R, Findlay L, Edwards C, Eastwood D, Bird C, North D, et al. Cytokine storm in the phase I trial of monoclonal mAb TGN1412: better understanding the causes to improve preclinical testing of immunotherapeutics. J Immunol 2007;179:3325–31. [12] Findlay L, Eastwood D, Stebbings R, Sharp G, Mistry Y, Ball C, et al. Improved in vitro methods to predict the in vivo toxicity in man of therapeutic monoclonal antibodies including TGN1412. J Immunol Methods 2010;352:1–12. [13] Wing M. Monoclonal mAb first dose cytokine release syndromes–mechanisms and prediction. J Immunotoxicol 2008;5:11–5. [14] Carpenter PA, Pavlovic S, Tso JY, Press OW, Gooley T, Yu XZ, et al. Non-Fc receptor-binding humanized anti-CD3 antibodies induce apoptosis of activated human T cells. J Immunol 2000;165:6205–13.

837

[15] Eastwood D, Findlay L, Poole S, Bird C, Wadhwa M, Moore M, et al. Monoclonal mAb TGN1412 trial failure explained by species differences in CD28 expression on CD4+ effector memory T-cells. Br J Pharmacol 2010;161: 512–26. [16] Römer PS, Berr S, Avota E, Na SY, Battaglia M, ten Berge I, et al. Preculture of PBMCs at high cell density increases sensitivity of T-cell responses, revealing cytokine release by CD28 superagonist TGN1412. Blood 2011;118: 6772–82. [17] Muller PY, Milton M, Lloyd P, Sims J, Brennan FR. The minimum anticipated biological effect level (MABEL) for selection of first human dose in clinical trials with monoclonal antibodies. Curr Opin Biotechnol 2009;20:722–9. [18] Langezaal I, Hoffmann S, Hartung T, Coecke S. Evaluation and prevalidation of an immunotoxicity test based on human whole-blood cytokine release. Altern Lab Anim 2002;30:581–95. [19] Walker MR, Makropoulos DA, Achuthanandam R, van Arsdell S, Bugelski PJ. Development of a human whole blood assay for prediction of cytokine release similar to anti-CD28 superagonists using multiplex cytokine and hierarchical cluster analysis. Int Immunopharmacol 2011;11:1697–705. [20] Dhir V, Fort M, Mahmood A, Higbee R, Warren W, Narayanan P, et al. A predictive biomimetic model of cytokine release induced by TGN1412 and other therapeutic monoclonal antibodies. J Immunotoxicol 2012;9:34–42. [21] Waibler Z, Sender LY, Kamp C, Müller-Berghaus J, Liedert B, Schneider CK, et al. Toward experimental assessment of receptor occupancy: TGN1412 revisited. J Allergy Clin Immunol 2008;122:890–2. [22] Waibler Z, Sender LY, Merten C, Hartig R, Kliche S, Gunzer M, et al. Signaling signatures and functional properties of anti-human CD28 superagonistic antibodies. PLoS One 2008;3:e1708. [23] Wing MG, Waldmann H, Isaacs J, Compston DA, Hale G. Ex vivo whole blood cultures for predicting cytokine-release syndrome: dependence on target antigen and mAb isotype. Ther Immunol 1995;2:183–90.