Impact of human sequences in variable domains of therapeutic antibodies on the location of CD4 T-cell epitopes

The clinical efficacy of therapeutic antibodies is recognized in many indications, explaining their rapid rise in the worldwide drug market. However, therapeutic antibodies can be immunogenic by inducing a specific immune response characterized by the production of anti-drug antibodies (ADAs). ADAs can potentially increase the clearance of therapeutic antibodies, decrease their therapeutic effects¹ and induce hypersensitivity reactions. Because humans are thought to be tolerant to their own proteins, sequences of the therapeutic antibodies have been humanized or human antibodies have been directly generated by technologies using human sequences. However, human(ized) antibodies exhibit highly variable levels of immunogenicity,^1,2 illustrating the poor understanding of the role of sequence humanization in mitigating the immunogenicity of therapeutic antibodies. Multiple lines of evidence highlight the role of CD4 T lymphocytes during the initiation of an ADA response,³ but few T-cell epitopes of marketed therapeutic antibodies have been described to date.

In this study, we identified T-cell epitopes from the human therapeutic antibody adalimumab (Adm) and the humanized antibody natalizumab (Ntz). Adalimumab is a human anti-human TNF-α antibody prescribed for the treatment of different inflammatory diseases and is known to induce ADA responses in many patients.¹ Natalizumab targets integrin α4β1 and prevents T lymphocytes from crossing the blood-brain barrier. Treatment with natalizumab efficiently reduces myelin loss in multiple sclerosis patients, but ADA responses are detected in multiple treated patients.

We identified CD4 T-cell epitopes of Adm and Ntz by generating T-cell lines with multiple stimulations of CD4 T-cells collected from healthy donors by autologous DCs loaded with either Adm or Ntz. With this approach, we isolated 28 independent T-cell lines specific for Adm peptides from 14 different donors covering a large diversity of HLA-DR allotypes (Fig. 1a, c). T-cell lines reacted specifically with 11 peptides clustered in three regions overlapping with HCDR2, HCDR3, and LCDR2. The two peptides from HCDR3 (AH91-110 and AH96-110) were recognized by a majority of T-cell lines isolated from 10 different donors. We isolated 10 independent T-cell lines specific for Ntz from six different donors out of 12 tested donors (Fig. 1b, c). These T-cell lines reacted with nine different peptides, localized in HCDR1, HCDR2, HFR3, LCDR1, and LCDR2. Only two peptides (NH51-65 and NH56-70) were common to two different donors (Fig. 1c).

Fig. 1.

Location of the CD4 T-cell epitopes in the Adm and Ntz VH and VL domains. Mab-specific T-cell lines were generated by 4 weekly rounds of stimulation of immunopurified CD4 T-cells collected from healthy donors with Adm (a) or Ntz (b) loaded on autologous dendritic cells. CD4 T-cell specificity was analyzed by IFN-γ ELISPOT assays using overlapping peptides covering the VH and VL domains of each Mab. Two sequential IFN-γ ELISPOT assays were performed using pools composed of 46 to 48 overlapping 15- and 20-mer peptides encompassing the VH and VL domains of the therapeutic antibodies in a first assay and individual peptides from these pools in a second assay. CD4 T-cell lines were considered specific when a spot count was two-fold higher than in the absence of peptides, with a minimal difference of 25 spots. Each T-cell line was found to be specific in two independent ELISPOT assays and was named by two numbers referring to the donor number and the culture well number. Examples of CD4 T-cell lines specific to a Adm peptides and to b Ntz peptides. c Sequence, HLA binding activity and T-cell reactivity of Adm and Ntz peptides are reported. CDR regions are shown in blue. Amino acids in orange correspond to mutations with respect to the best-fitting germline sequence based on IMGT analysis.⁷ The best-fitting germline sequences of Adm are IGHV3-9*01, IGHJ4*02, IGHD2-2*01, IGKV1-27*01 and IGKJ2*01 and are IGHV1-3*01, IGHJ6*01, IGKV1-33*01, and IGKJ1*01 for Ntz. Amino acids highlighted in yellow correspond to mutated positions relative to the best-fitting germline sequence but present in at least one 9-mer peptide from the human variable domain according to the Human String Content (HSC) method.⁶ Binding activity to common HLA-DR molecules was evaluated by competitive ELISA specific for molecules encoded by the HLA-DRB1 gene (01:01, 03:01, 04:01, 07:01, 11:01 and 15:01) and 2^nd HLA-DR molecules (DRB4*01:01 and DRB5*01:01). The binding activity of Ntz peptides was also assessed for DRB1*09:01 and DRB3*01:01. The number of bound HLA-DR molecules is reported for peptides bearing mutations with respect to the best-fitting germline sequence

We applied the MHC-associated peptide proteomic (MAPP) approach to a panel of donors to identify Adm and Ntz peptides displayed by HLA-DR molecules on DCs. HLA-DR-associated peptides were isolated from Mab-loaded DCs and identified by liquid chromatography-mass spectrometry. Identified peptides were clustered in multiple sets of length variants sharing the same binding cores, and multiple clusters overlapped by at least nine amino acids with the CD4 T-cell epitopes (supplementary Table 3).

We also assessed the capacity of Adm and Ntz peptides to bind to a selection of common HLA-DR alleles in the worldwide population (Fig. 1c). The T-cell epitopes AH46-60, AH51-65, and AH76-90 did not show good affinity to the HLA-DR alleles tested here but were still found by MAPPs, confirming the capacity of these peptides to be presented by HLA-DR molecules. In contrast, the Adm heavy-chain region 86 to 120, which was a hotspot of good binders to HLA-DR molecules (Fig. 1c), contained the vast majority of identified T-cell epitopes. Almost all the Ntz peptides exhibited a good or moderate affinity for HLA-DR molecules (Fig. 1c). We also noticed the lack of T-cell-stimulating peptides arising from the HCDR3 of Ntz, along with the lack of presented peptides in this region according to MAPPs, although this region contained multiple good binders to HLA-DR molecules.

Clearly, the locations of the identified CD4 T-cell epitopes of Adm and Ntz are different and differed from those of the chimeric antibodies analyzed in a previous study,³ demonstrating the uniqueness of a T-cell response against therapeutic antibodies. For both Adm and Ntz antibodies, CDR regions appeared to be the main driver of the T-cell response, as previously suggested.⁴ Studies in mice demonstrated that the somatic hypermutation process created CD4 T-cell epitopes in V regions, although germline sequences did not elicit a CD4 T-cell response.⁵ Germline sequences lead to central deletion of the specific CD4 T-cell precursors in the thymus, but some of the T-cells may evade thymic selection.⁵ In our study, sequences of only three Adm T-cell epitopes (AH76-90, AH86-100 and AL46-60) and one Ntz T-cell epitope (NH71-95) were found to be nonmutated with respect to germline sequences.⁶ The majority of T-cell epitopes of both antibodies identified in our study are highly mutated. The CDR3 sequences result from the junction of V(D)J segments and TdT-catalyzed addition of nucleotides; therefore, the CDR3 sequences largely differ from germline sequences. In Adm, this may contribute to rendering the HCDR3 a T-cell epitope hotspot and is further supported by the efficient binding and presentation properties of this part of the Adm sequence. In CDR1 and CDR2 of the VH and VL chains, somatic mutations are introduced during the affinity maturation process. In our study, T-cell responses were also observed against Ntz HCDR1, HCDR2, LCDR1, LCDR2, and Adm HCDR2 and LCDR2, which all contained somatic mutations. Together, our data strongly suggest that the presence of mutations in a peptide combined with its capacity to bind to HLA-DR and to be presented by DCs largely contributes to the human CD4 T-cell response to both tested human(ized) antibodies. This concept is generally anticipated in the design of new therapeutic antibodies but has never been formally demonstrated. Furthermore, we have shown that the degree of humanness per se is not sufficient to discriminate antibodies on the basis of their T-cell stimulatory capacity. Adm induced markedly more T-cells than Ntz, although Adm is closer to germline sequences than Ntz. Despite efforts to generate antibodies with human(ized) frameworks, the nongermline amino acid sequences mainly found in the CDRs might continue to drive immunogenicity against monoclonal antibodies.

Author contributions

S.M., M.H., A.K., M.B., A.G., S.S., and B.M. designed the experiments; S.M., M.H., M.B., and A.G. performed the experiments; S.M., M.H., A.K., M.B., A.G., S.S., and B.M. analyzed the data; and A.G., S.S., and B.M. wrote the paper.

Competing interests

A.K. is full-time employee of and holds shares and stock options in Novartis. S.S. is a former employee and has stocks and stock options in Novartis.

Supplementary information

The online version of this article (10.1038/s41423-019-0304-3) contains supplementary material.