Specificity of human anti-variable heavy (V_H) chain autoantibodies and impact on the design and clinical testing of a V_H domain antibody antagonist of tumour necrosis factor-α receptor 1

Abstract

During clinical trials of a tumour necrosis factor (TNF)-R1 domain antibody (dAb™) antagonist (GSK1995057), infusion reactions consistent with cytokine release were observed in healthy subjects with high levels of a novel, pre-existing human anti-V_H (HAVH) autoantibody. In the presence of HAVH autoantibodies, GSK1995057 induced cytokine release in vitro due to binding of HAVH autoantibodies to a framework region of the dAb. The epitope on GSK1995057 was characterized and dAbs with reduced binding to HAVH autoantibodies were generated; pharmacological comparability was determined in human in-vitro systems and in-vivo animal experiments. A Phase I clinical trial was conducted to investigate the safety and tolerability of the modified dAb (GSK2862277). A significant reduction in HAVH binding was achieved by adding a single alanine residue at the C-terminus to create GSK2862277. Screening a pool of healthy donors demonstrated a reduced frequency of pre-existing autoantibodies from 51% to 7%; in all other respects, GSK2862277 and the parent dAb were comparable. In the Phase I trial, GSK2862277 was well tolerated by both the inhaled and intravenous routes. One subject experienced a mild infusion reaction with cytokine release following intravenous dosing. Subsequently, this subject was found to have high levels of a novel pre-existing antibody specific to the extended C-terminus of GSK2862277. Despite the reduced binding of GSK2862277 to pre-existing HAVH autoantibodies, adverse effects associated with the presence of a novel pre-existing antibody response specific to the modified dAb framework were identified and highlight the challenge of developing biological antagonists to this class of receptor.

Introduction

Domain antibodies (dAbs) are the smallest functional binding units of human immunoglobulin (Ig)G antibodies with a molecular weight of between 10–13 kDa, and are derived from fully human sequences of the variable regions of either the heavy (V_H) or light (V_L) chains. We have developed V_H dAbs that potently and selectively antagonize binding of tumour necrosis factor (TNF)-α to TNF receptor 1 (TNF-R1) and attenuate inflammation and cell death in animal models of acute lung injury 1. During Phase I clinical trials of a first-generation TNF-R1 dAb antagonist (GSK1995057) in healthy human subjects, pre-existing, naturally occurring anti-immunoglobulin autoantibodies were discovered in the serum of approximately 50% of drug-naive, normal humans 2. These pre-existing anti-drug antibodies (ADA) were shown to bind to an epitope close to the C-terminal region of V_H dAbs, including GSK1995057. In the case of GSK1995057, the complex of these endogenous non-neutralizing human anti-V_H (HAVH) autoantibodies with framework sequences of GSK1995057 led to activation of TNF-R1 signalling 2 and the occurrence of mild–moderate infusion reactions in subjects with high HAVH autoantibody titres.

Here we describe studies that further identify HAVH binding epitopes within the C-terminus of GSK1995057 and, through mutagenesis experiments, the generation of TNF-R1-binding dAbs with reduced binding to HAVH autoantibodies and a lower risk of autoantibody-mediated receptor activation. Moreover, we have found that a single amino acid extension of the C-terminal framework sequence is sufficient to reduce binding of GSK1995057 to HAVH autoantibodies, while retaining comparable pharmacology and biophysical properties. Reduced binding to HAVH autoantibodies was demonstrated across a panel of human sera, and in a functional assay of TNF-R1 signalling and pharmacological comparability was demonstrated in human in-vitro systems and animal experiments in vivo. Finally, a Phase I clinical trial was conducted to investigate the safety and tolerability of single and repeat intravenous (i.v.) and inhaled (i.h.) doses of the modified dAb (GSK2862277) in healthy adult subjects with HAVH autoantibodies. The study also investigated the comparability of pharmacokinetics (PK) and pharmacodynamics (PD) with historical clinical data generated with the parent dAb, GSK1995057.

Materials and methods

Reagents

GSK1995057 and its modified variants and recombinant cynomolgus monkey TNF-R1 : Fc fusion protein were produced by GlaxoSmithKline (Stevenage, UK). MRC-5 and A549 human lung cell lines originated from the ATCC® (CCL-171 and CCL-185, respectively). CYNOM-K1 cynomolgus monkey skin fibroblast cell line originated from European Collection of Cell Cultures (ECACC, Salisbury, UK) (#90071809). Recombinant human TNF-α was supplied by Peprotech (Rocky Hill, NJ, USA) and recombinant human TNF-R1 : Fc fusion by R&D Systems (Minneapolis, MN, USA).

Mutagenesis

The GSK1995057 expression vector was used as a template to generate a range of modified variants of GSK1995057 by polymerase chain reaction (PCR) using specific primers and a QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA, USA). Comparison of the GSK1995057 framework sequence to different single-domain antibodies that either competed or not for GSK1995057 HAVH autoantibody binding 2 identified residues that might be involved in HAVH binding. Single- or double-framework mutations (changed to alanine) of these residues and selected other positions in close proximity were generated as well as C-terminal extensions to GSK1995057. The identity of all expression constructs was confirmed by sequencing and dAbs expressed from Escherichia coli. A range of in-vitro and in-vivo assessments were carried out to determine the comparability of GSK1995057 and GSK2862277. Further details of these analyses are presented in the Supporting information.

Assay for cross-reactivity of autoantibodies with GSK1995057 and GSK2862277

An electrochemiluminescent (ECL) antibody detection assay was used to measure autoantibodies that bind to GSK1995057 in human serum samples, as described previously 2. The mean ECL signal from sample replicates was divided by the mean ECL signal from negative control replicates to give a relative ECL (RECL) value. Assay cut-points were determined as described previously 2. To assay for cross-reactivity of modified dAbs with HAVH autoantibodies, human sera were preincubated with an excess of modified dAb prior to incubation with biotinylated and SULFO-TAG™-labeled GSK1995057. Inhibition of the ECL signal indicates cross-reactivity of the modified dAb with the HAVH autoantibodies.

Clinical trial design

The study (TFR116343) was a three-part, randomized, placebo-controlled Phase I clinical trial to investigate the safety, tolerability, pharmacokinetics and pharmacodynamics of single and repeat doses of inhaled or intravenous GSK2862277 in healthy volunteers (ClinicalTrials.gov identifier NCT01818024). The design of the study is depicted in Fig. 1. Healthy males or females between the ages of 18 and 65 years were eligible to enrol in this study. Females were required to be of non-child-bearing potential.

Figure 1.

Study schematic. A = active (GSK2862277); P = placebo; filled and empty diamonds represent interim decision points based on safety.

In part 1 of the study, subjects that had been prospectively screened positive for HAVH antibodies to the previously tested dAb GSK1995057 were given ascending i.v. doses of GSK2862277 up to 0·05 mg/kg. Part 1 was conducted in an open-label fashion without inclusion of placebo cohorts. Previous clinical investigations with GSK1995057 incorporated placebo subjects over a full-dose escalation design with almost identical eligibility criteria 2, and so were of use in interpreting potential idiosyncratic toxicities. The absence of cytokine release in HAVH autoantibody-positive subjects dosed with GSK2862277 in part 1 would support non-clinical data showing that GSK2862277 has reduced binding to HAVH autoantibodies and does not induce activation of TNF-R1 signalling, and thus cytokine release in humans. The lack of any evidence of cytokine release in dosed subjects in part 1 would enable progression of the study without the need for prospective identification of a subject's HAVH status.

Part 2 of the study investigated single i.v. (2 mg/kg) or i.h. (26 mg) doses of GSK2862277, while part 3 of the study investigated repeat doses (once daily for 5 days) of i.v. (2 mg/kg) or i.h. (26 mg) doses. Subjects in parts 2 and 3 were not tested for HAVH autoantibodies and assigned to GSK2862277 or matching placebo in accordance with a randomization schedule generated by GSK Biopharm Statistics. Randomization schedules were also generated to assign subjects to intravenous or inhalation cohorts in part 3.

TNF-α induced interleukin (IL)−8 release from whole blood following i.v. dosing with GSK2862277 was measured as a biomarker of pharmacodynamic activity (TNF-R1 inhibition). Blood was collected into TruCulture® tubes (Myriad RBM, Reutlingen, Germany) prefilled with 10 ng/ml TNF-α (R&D Systems). After incubation for 24 h, samples were processed following the manufacturer's instructions and IL-8 levels quantified using commercial assays from Immulite® (Siemens Healthcare, Erlangen, Germany). TNF-α and IL-6 levels in serum were also measured using Immulite assays.

Samples for immunogenicity testing were collected at baseline (day −1) up to day 60 and tested as described 2. The titre, or the highest dilution factor that still yielded a positive reading for the sample, was reported for each positive sample. The titre value also incorporated an assay dilution factor of 1 : 50, therefore the minimum measurable titre was 50.

The study was approved by the Edinburgh Independent Ethics Committee for Medical Research and conducted at Parexel International Early Phase Clinical Unit (Northwick Park Hospital, Harrow, UK UK). The study was conducted in accordance with the revised Declaration of Helsinki (2008), good clinical practice and all applicable regulatory requirements.

Statistical analysis

A sample size of 56 subjects was planned to complete dosing and study procedures relating to the primary end-point of safety and PK, based primarily on feasibility. There were no formal interim analyses performed; however, data reviews were conducted on an in-stream basis both within and between study parts. The primary safety end-points were evaluated based on descriptive statistics.

Pharmacokinetic parameters were calculated by standard non-compartmental analysis using Win Nonlin Pro version 5·2 or higher. All calculations of non-compartmental parameters were based on actual sampling times. A comparability assessment between GSK2862277 and GSK1995057 was performed to confirm a similar GSK2862277 PK profile to that obtained from prior investigations with GSK1995057. Further details of these analyses are presented in the Supporting information.

Results

Identification and characterization of GSK2862277

Identification of epitopes with reduced ability to bind HAVH autoantibodies

A comparison of the GSK1995057 framework sequence to different single domain antibodies that either competed or not for GSK1995057 HAVH autoantibody binding 2 was undertaken to identify and mutate specific amino acid residues that were predicted to be involved in HAVH autoantibody binding. Single and pairwise mutations were introduced at these residues and at selected other positions in close proximity (Fig. 2). Additionally, the C-terminus was extended by one residue at a time using naturally occurring CH1 domain residues of human IgG, thereby reducing the risk of the extensions themselves being immunogenic.

Figure 2.

C-terminal extensions have reduced ability to inhibit human anti-variable region immunoglobulin (Ig)G heavy chain (VH) (HAVH) autoantibody binding. Binding of HAVH autoantibodies to GSK1995057 is inhibited by preincubation of HAVH-positive serum samples with free unlabelled GSK1995057 and to a varying degree by preincubation with mutated domain antibodies (dAbs). Variants with a P14A substitution or with C-terminal extensions up to +ASTKGP have reduced ability to inhibit HAVH autoantibody binding to GSK1995057. Horizontal line indicates assay cut-point for cross-reactivity with HAVH autoantibodies.

The modified and extended versions of GSK1995057 were tested for their ability to reduce binding of HAVH autoantibodies to GSK1995057 (Fig. 2) and the results mapped onto an X-ray crystallographic structure (unpublished) of GSK1995057 (Fig. 3). The most notable reduction in binding was achieved by substituting the proline at position 14 (Kabat numbering) 3 with alanine, and also in molecules with C-terminal extensions derived from the start of the CH1 domain, ranging from a single alanine up to and including the 6 amino acid extension, ‘ASTKGP’. The molecule with a single alanine extension at the C-terminus, designated GSK2862277, represents the most conservative sequence change that results in reduced HAVH autoantibody binding, and was therefore selected for further study.

Figure 3.

In-silico identification of residues in GSK1995057 which may be involved in binding of HAVH autoantibodies. Green = no predicted involvement, yellow = predicted weak involvement, light red = predicted moderate involvement, dark red = predicted strong involvement. Residues are annotated using the Kabat numbering scheme.

Frequency of autoantibodies to GSK2862277 is reduced compared with GSK1995057

A panel of 100 healthy donor human serum samples that had been screened previously for the presence of HAVH autoantibodies that bind GSK1995057 were screened using a similar assay format for the presence of autoantibodies that bind GSK2862277 (Fig. 4a). The results show that the frequency of GSK2862277-specific autoantibodies was much reduced (7% of samples) compared with GSK1995057-specific HAVH autoantibodies (51% of samples). This finding was confirmed by screening a second serum panel from 100 healthy donors in which 6% of samples contained antibodies that bound to GSK2862277 (data not shown). There was no evidence of a correlation between the presence of autoantibodies to GSK1995057 and GSK2862277 in the same donors (Fig. 4b), although the subject with the highest level of GSK2862277-specific autoantibodies also tested positive for GSK1995057-specific HAVH autoantibodies.

Figure 4.

Screening for autoantibodies. (a) A panel of human serum samples was screened in assays designed to detect autoantibodies specific for GSK1995057 or GSK2862277 (horizontal line represents assay cut point). (b) Correlation analysis shows no significant correlation between autoantibodies specific for GSK1995057 and GSK2862277 in human serum.

Effect of autoantibodies on the pharmacology of GSK1995057 and GSK2862277

As reported previously 2, we have developed an in-vitro cell-based assay system for assessing GSK1995057-dependent, HAVH autoantibody-mediated TNF-R1 activation (IL-8 release) using a human lung fibroblast cell line. To investigate whether the single alanine extension would reduce TNF-R1 activation in the presence of HAVH autoantibodies, GSK2862277 was mixed with HAVH autoantibody-positive human serum samples shown previously to mediate activation of TNF-R1 in the presence of GSK1995057 2. No evidence of TNF-R1 activation was observed when GSK2862277 was tested in this assay (Fig. 5a).

Figure 5.

In-vitro activation of tumour necrosis factor (TNF)-R1. (a) GSK2862277 was incubated with human sera that contain HAVH autoantibodies (solid lines) or those that test negative (broken lines). In-vitro activation of TNF-R1 by antibody complexes was measured by interleukin (IL)−8 release from MRC-5 cells. (b,c) Serum samples which had tested positive (solid lines) or negative (broken lines) for autoantibodies specifically to GSK2862277 were tested for TNF-R1 activation when preincubated with GSK1995057 (b) or GSK2862277 (c). Points represent mean of assay replicates ± standard deviation.

To investigate the possibility that de-novo pre-existing GSK2862277-specific autoantibodies could induce TNF-R1 activation in the same way as seen for GSK1995057, GSK1995057 or GSK2862277 were mixed with human serum samples which had tested positive for GSK2862277-specific autoantibodies before being tested in the in-vitro TNF-R1 activation assay. A number of these sera were also positive for GSK1995057-specific HAVH autoantibodies and therefore induced TNF-R1 activation in the in-vitro assay when complexed with GSK1995057 (Fig. 5b). However, no evidence of TNF-R1 activation was observed with GSK2862277 in combination with any of the human sera tested, including in samples with GSK2862277-specific autoantibodies (Fig. 5c).

Comparability of GSK1995057 with GSK2862277

A number of studies were conducted to demonstrate comparability of GSK2862277 to GSK1995057. These studies included surface plasmon resonance to compare binding affinity for human and cynomolgus monkey TNF-R1, in-vitro inhibition of TNF-R1 signalling in human and cynomolgus monkey cell lines and whole blood, cynomolgus monkey PK and efficacy in a lipopolysaccharide model of acute lung inflammation in cynomolgus monkeys. In all respects GSK2862277 and GSK1995057 were found to be highly comparable in terms of target affinity, in-vitro potency and in-vivo pharmacokinetics and pharmacodynamics in non-human primate models (Supporting information, Table S1).

Phase I clinical trial with GSK2862277

Safety

Fifty-six subjects were randomized into the trial; subject disposition and demographics are outlined in Table1. Nineteen (56%) of the 34 healthy subjects who were administered single doses of GSK2862277 (parts 1 and 2) experienced at least one adverse event (AE) (Supporting information, Table S2). The most common AE in the single-dose period was headache; other individual AEs in the single-dose period occurred in one or two subjects each, with no dose-related pattern for any of the AEs during the single-dose period.

Table 1.

Subject disposition

	GSK2862277	Placebo
Number of subjects
Planned, n	46	10
Randomized, n	46	10
Completed, n (%)	46 (100)	10 (100)
Total number subjects withdrawn, n (%)	0 (0)	0 (0)
Male, n (%)	46 (100)	10 (0)
Race
White, %	77
Asian, %	13
African American/African heritage, %	11

In the repeat-dose period of the study (part 3), nine (56%) of the 16 subjects who were administered repeat doses experienced at least one AE (Supporting information, Table S3). Adverse events were reported more commonly in the GSK2862277 treatment groups than placebo for both i.v. and i.h. dosing. The most common AE in the repeat-dose period was nasopharyngitis, reported for three subjects (50%) in the GSK2862277 26 mg i.h. dose group.

One subject (who received daily 2-mg/kg i.v. doses of GSK2862277 for 5 days) presented on day 6 (24 h after administration of the last infusion) with mild myalgia, fatigue and slightly raised temperature (maximum temperature 38·3°C) accompanied by a rise in serum C-reactive protein (CRP) (maximum 181·6 mg/l), raised transaminases (ALT max = 193 IU/l and AST max = 94 IU/l) and low lymphocyte count (nadir of 0·38 GI/l). Clinical signs resolved rapidly and laboratory abnormalities returned to normal levels in 4 days; viral serologies were negative for this subject. Subsequently, this subject was shown to have very high serum levels of pre-existing antibodies to GSK2862277 that remained stable throughout the dosing and follow-up periods.

Increases in serum cytokines were observed shortly after initiation of dosing and then contemporaneously with the occurrence of clinical signs on day 6. Following the initiation of dosing, TNF-α levels were slightly elevated above the upper limit of what is considered normal for a healthy subject (Fig. 6) before returning to baseline levels by the 24-h sample time-point and remaining within the normal range throughout the remainder of the study. However, in addition to a small peak in IL-6 levels at 25 h following start of dosing, levels increased significantly on day 6, returning to within the normal range by day 7. The ability of serum autoantibodies from this subject to activate TNF receptor signalling was confirmed using the previously described in-vitro assay.

Figure 6.

Cytokine release in a patient with elevated titres of anti-drug antibodies (ADAs). Plasma levels of interleukin (IL)−6, tumour necrosis factor (TNF)-α (left axis) and C-reactive protein (CRP) (right axis) were measured over time following intravenous dosing with GSK2862277 2 mg/kg once daily for 5 days.

With the exception of the subject described above, there were no clinically significant changes in any laboratory safety parameters, electrocardiography (ECG), vital signs or spirometry. Serum TNF-α levels remained relatively stable and consistent with predose levels. While IL-6 levels were more variable, they generally remained within normal levels (e.g. < 10 pg/ml 4) over the duration of the study.

Pharmacokinetics

A dose-proportional increase in plasma GSK2862277 was observed between i.v. administration of 0·002 mg/kg and 0·01 mg/kg; however, plasma exposures increased more than dose-proportionally as the dose increased to 0·05 mg/kg and 2 mg/kg (Supporting information, Fig. S1). The plasma half-life was calculated at 2·9–5·5 h [90% confidence interval (CI)] and 9·6–17 h (90% CI) for i.v. and i.h. doses, respectively. Steady state was reached by day 5 and there was no accumulation following 5 days repeat i.v. or i.h. dose administration of GSK2862277 (data not shown). The fractional excretion and renal clearance could not be analysed due to an analytical failure; however, the urine half-life was calculated to be 4·0 h (90% CI = 2·9–5·4).

Ex-vivo IL-8

Effects on TNF-R1 signalling were measured by inhibition of ex-vivo TNF-α-induced IL-8 release in whole blood. Figure 7a shows the relationship between GSK2862277 concentrations and ex-vivo TNF-α-stimulated IL-8 levels in the blood of subjects dosed with 0·5 and 2 mg/kg i.v. GSK2862277. While IL-8 release was variable across subjects, a clear reduction is apparent in blood samples collected when high plasma concentrations of GSK2862277 were recorded. To compare the PK/PD relationship of GSK2862277 to the parent molecule GSK1995057, ex-vivo-stimulated IL-8 data and exposure data in healthy subjects for both molecules were merged into a single PK/PD plot (Fig. 7b). Although no formal statistical analysis was performed on this merged data set, the graphical analysis suggests GSK1995057 and GSK2862277 exhibit a comparable relationship between PK and PD in humans.

Figure 7.

GSK2862277 inhibits ex-vivo tumour necrosis factor (TNF)-α-induced interleukin (IL)-8 release in whole blood. (a) Ex-vivo TNF-α-induced IL-8 was measured as described in the Methods. TNF-α-induced IL-8 levels in whole blood from predose samples ranged from 339 to 2094pg/ml in placebo subjects (n = 4) and from 311 to 3268pg/ml in subjects dosed subsequently with GSK2862277 (n = 18). (b) Comparison of ex-vivo IL-8 data from GSK1995057 and GSK2862277. Red vertical dashed lines = in-vitro IC₅₀ and IC₉₀ parameters for GSK2862277 and GSK1995057. Black vertical dashed lines = lower limit of quantification (LLQ) of GSK2862277 and GSK1995057. Black horizontal dashed line = LLQ of IL-8.

Immunogenicity

Eleven subjects tested positive for anti-GSK2862277 antibodies at some time-point during the study; framework-specific antibodies were detected in 10 of the 11 positive subjects. Seven of these 11 subjects were positive for pre-existing anti-GSK2862277 antibodies, four of whom had titres above 200 (Supporting information, Table S4).

Among these four was the subject who experienced adverse events related to cytokine release following i.v. dosing, and who had a baseline predose antibody titre of 49,000 (corresponding to a RECL value of 83·26). The antibody titre remained high throughout the study period (peak titre of 63,980, RECL value of 145·95 at day 14; titre of 41,460 RECL value of 89·96 on day 60). Predose serum from this subject induced IL-8 release in the in-vitro TNF-R1 activation assay when complexed with GSK2862277 (Fig. 8). Sera from the other three subjects with high predose levels of anti-GSK2862277 antibodies did not induce TNF-R1 activation in the in-vitro assay (data not shown).

Figure 8.

Pre-existing GSK2862277-specific autoantibodies activate tumour necrosis factor (TNF)-R1 in vitro. Predose serum from a subject with a high titre for GSK2862277-specific antibodies was incubated with increasing concentrations of GSK2862277 and tested for TNF-R1 activation in vitro (black solid lines). For comparison, serum obtained pre- and post-dose from a subject who tested negative for pre-existing GSK2862277-specific autoantibodies but who seroconverted post-dosing was also tested (grey broken lines). Points represent mean of assay replicates ± standard deviation.

Four subjects seroconverted during the dosing period. Two of these subjects had titre values less than twofold above the minimum measurable level by day 7, and negative titres by day 60; a third subject only had measurable low-level titres at day 60. The fourth subject, who received 26 mg GSK2862277 inhaled once daily for 5 days, was found to have the most elevated post-dose titres, but did not experience any adverse events related to cytokine release. This subject had low-level titres on day 3, a peak titre of 5060 (RECL = 4·41) on day 14 and a titre of 476 (RECL = 1·44) at day 60. The antibody was identified as an IgM isotype. Neither pre- nor post-dose sera from this subject induced IL-8 release in the in-vitro TNF-R1 activation assay when complexed with GSK2862277 (Fig. 8).

Discussion

In previous work, we identified the presence of pre-existing HAVH autoantibodies to the V_H framework of a fully human TNF-R1 domain antibody antagonist GSK1995057. In human subjects with high levels of these pre-existing HAVH antibodies, administration of GSK1995057 resulted in mild–moderate, transient infusion reactions accompanied by cytokine release at certain dose levels. Interactions between pre-existing HAVH antibodies and a DR5-targeting V_HH camelid domain antibody (TAS266) have also been reported recently that resulted in transient and reversible liver toxicity in a Phase I clinical trial in healthy subjects 5. Consistent with our experience with GSK1995057, pre-existing anti-TAS266 antibodies were also present in Cynomolgous monkeys, but did not alter pharmacology or result in toxicity in this species.

Anti-IgG autoantibodies are common in human subjects 6–8, although their origin and function are largely unknown. It has been suggested that they may arise as part of an adaptive immune response to fragments resulting from normal or pathological processing of IgG 9–11 and that their roles may include immunoregulation 9, host defence 12 or clearance of antibodies, apoptotic cells and other waste materials 10,13. It is notable that HAVH antibodies bind a cryptic epitope at the C-terminal epitope of V_H dAbs, which is not naturally accessible to HAVH antibodies in full IgG molecules 2. This further raises the possibility that this epitope is revealed within IgG fragments or metabolites following either normal or pathological IgG catabolism.

Analysis of existing data on biological therapeutics in clinical development suggests that pre-existing anti-drug antibodies are encountered widely 14,15, and adverse events attributable to pre-existing anti-drug antibodies have been reported in human subjects following dosing with biotherapeutics 16,17, highlighting the importance of further understanding pre-existing anti-drug antibodies and their potential implications for drug safety profiles. Given the adverse effects associated with dosing GSK1995057 in humans with high levels of HAVH, we undertook studies to identify amino acid residues proximal to the common C-terminal framework region of V_H dAbs that might be responsible for binding to HAVH antibodies. Furthermore, using a site-directed mutagenesis approach, we explored whether binding of HAVH antibodies to V_H dAbs could be reduced through specific amino acid substitutions. These experiments confirmed the importance of residues at, or conformationally proximal to, the C-terminus of the V_H dAb common framework region for binding HAVH antibodies. Indeed the most dramatic impact on HAVH antibody binding was observed with either truncated or extended C-terminal variants. A simple C-terminal alanine extension reduced binding of HAVH antibodies to GSK1995057 substantially, suggesting a prevailing clonal response to this C-terminal V_H epitope among the HAVH antibody repertoire. Linear peptides comprising the 11 C-terminal amino acids of GSK1995057 were also tested and had no effect on HAVH antibody binding, suggesting a conformational rather than a linear epitope (data not shown).

The single alanine extension of the C-terminus was selected as the most conservative change to GSK1995057 that resulted in reduced HAVH antibody binding but retained equivalent potency and biophysical properties compared with the parental dAb. In addition, as the alanine is normally present at this position in natural human IgG, being the first residue of the CH1 domain (position 118 by Eu numbering) 18, there was less likelihood of introducing an additional immunogenic motif. Importantly, there was a significantly lower frequency of pre-existing antibodies specific to GSK2862277, as assessed by screening healthy volunteer serum panels, and there was no evidence of in-vitro TNF-R1 activation following incubation of GSK2862277 with GSK1995057-specific HAVH-positive human sera.

With respect to target affinity, potency and in-vivo PK and PD, GSK1995057 and GSK2862277 were shown to be highly comparable. This is consistent with the location of the C-terminus within a common framework region that is not involved in target binding and is distant to the three antigen binding loops of the dAb. Because absorption and distribution of dAbs in vivo is governed predominantly by a combination of molecular weight (size) and target pharmacology (receptor binding), a single amino acid extension that does not alter target pharmacology and does not increase the size of the dAb significantly was not expected to alter pharmacokinetics or the primary elimination route of the dAb (renal filtration) in vivo.

To validate our hypothesis that reduced HAVH antibody binding to the modified dAb (GSK2862277) would result in a lower risk of infusion reactions and cytokine release in humans while maintaining comparable clinical pharmacology, we initiated a carefully designed Phase I trial in healthy human subjects. In the first part of the study, slowly escalating single i.v. doses of GSK2862277 were administered to healthy subjects who prospectively screened positive for high serum levels of GSK1995057-specific HAVH antibodies. Dosing in this phase was initiated from very low levels and progressed slowly into a dose range in which mild–moderate infusion reactions were observed with the previous dAb, GSK1995057, in subjects with high HAVH antibodies. In the first part of the study there was no evidence of infusion reactions or cytokine release in any subject over a broad range of intravenous GSK2862277 dose levels, regardless of HAVH autoantibody level. These data support our hypothesis that modification (extension) of the C-terminus of the parent dAb GSK1995057 reduces binding of HAVH antibodies associated previously with adverse effects in subjects dosed with GSK1995057. These human data also translate well from the in-vitro TNF-R1 activation assay data used to characterize GSK2862277 prior to initiation of Phase I trials.

The clinical pharmacology of GSK2862277 was observed to be within an acceptable range based on predicted PK parameters, with a 5-h half-life following 2 mg/kg i.v. administration and an approximately 12·5-h half-life following i.h. administration. A formal statistical assessment showed the clinical pharmacology of GSK2862277 to be comparable to historical data in healthy subjects dosed with the parent molecule GSK1995057. Furthermore, pharmacodynamic activity of GSK2862277 in blood taken from dosed human subjects was also comparable to historical clinical trial data with GSK1995057.

In the subsequent parts of the trial no preselection of subjects was made based on pre-existing antibody status, and the clinical pharmacology of GSK2862277 was assessed at therapeutic dose levels and following repeat dosing for 5 days. While GSK2862277 was generally well tolerated in parts 2 and 3 of the trial, one subject experienced a mild infusion reaction with concomitant cytokine release, and was found to have very high levels of pre-existing serum antibodies to specific to GSK2862277. Interestingly, the antibody response in this subject was found not to bind the parent molecule GSK1995057, and so was distinct in specificity from previous HAVH antibodies shown to bind and interact with GSK1995057. In the subjects with pre-existing ADAs, there was no evidence that treatment with GSK2862277 resulted in an anamnestic response, suggesting that the dAb itself did not augment ADA reactivity.

Despite the low incidence of pre-existing antibodies to GSK2862277 determined in 200 human serum samples screened prior to initiating clinical trials, adverse events in the one subject probably occurred due to an interaction of GSK2862277 with pre-existing autoantibodies that bind the new (‘+A’) C-terminus created in GSK2862277. It seems likely that this interaction resulted in antibody-mediated, GSK2862277-dependent cross-linking of cellular TNF-R1 and subsequent cytokine release, in a manner similar to that seen previously with GSK1995057. Symptom onset did not occur until 24 h after the last (5th) daily administration of 2 mg/kg i.v. GSK2862277, in contrast to previous clinical experience with GSK1995057. This seemingly counterintuitive observation may be explained by high concentrations of free (unbound) GSK2862277 in the plasma following administration of a 2 mg/kg dose that may preferentially saturate TNF-R1 and block the binding of any pre-existing antibody–dAb complexes. As the dAb is cleared from the body, GSK2862277 concentrations in the plasma reach a point at which antibody–dAb complexes are in excess over free GSK2862277 and can then bind and activate free TNF-R1 on target cells. The bell-shaped relationship between the concentration of GSK2862277 and the level of TNF-R1 activation, seen when serum from this subject was tested in the in-vitro TNF-R1 activation assay, also supports this mechanistic explanation and is similar to that seen for HAVH antibodies in combination with GSK1995057 2. The level of pre-existing anti-GSK2862277 antibodies in this subject was much higher (RECL 82·36) than any of the human serum samples screened and tested in the in-vitro TNF-R1 activation assay prior to the clinical study, where no RECL values for GSK2862277 framework-specific antibodies were greater than 3·5.

Finally, it is notable that, despite the high dose of GSK2862277 administered to this subject in the presence of very high levels of pre-existing antibodies, the infusion reaction seemed to be mild and self-limiting. This is consistent with previous infusion reactions observed in a few subjects with high pre-existing HAVH antibody levels who received i.v. GSK1995057 2 which were also self-limiting and mild–moderate in severity. Given the small number of anti-TNF-R1 dAb-related infusion reactions observed in humans to date, it is difficult to predict whether severity may vary and which factors are most important in controlling the severity of the reaction. However, it is possible that the complex of TNF-R1–targeting dAbs and pre-existing antibodies may form receptor (TNF-R1) dimers that signal submaximally relative to the normal formation of homotrimeric TNF : TNF-R1 ‘super-clusters’ in response to high natural ligand concentrations.

In summary, GSK2862277, a TNF-R1 dAb engineered for reduced binding to pre-existing antibodies, demonstrated improved tolerability following both inhaled and intravenous administration compared with a parent molecule, GSK1995057. Despite this improved profile, adverse effects associated with high levels of pre-existing antibodies specific to the modified dAb framework in GSK2862277 were identified in one subject and highlight the challenge of developing biological antagonists to this class of receptor. Further clinical trials of GSK2862277 are planned in patients confirmed negative for pre-existing ADAs, while investigations into pre-existing antibodies to GSK2862277 and the relationship between titre levels and risk of infusion reactions continues.

Author contributions

P. J. B., R. W., J. T., Y. C., T. J. W. and A. I. B. designed the clinical study; J. C. C., P. J. M., A. P. L., R. E. and M. A. B. designed and performed experiments; all authors analysed and/or interpreted the data; J. C. C., P. J. M., T. J. W., A. I. B. and A. L. L. wrote the paper; all authors reviewed the paper.

Disclosure

All authors are employees of GlaxoSmithKline and hold stock in the company.

Supporting Information

Additional Supporting information may be found in the online version of this article at the publisher's web-site: