Development of the clinical candidate PBD-C06, a humanized pGlu3-Aβ-specific antibody against Alzheimer’s disease with reduced complement activation

In clinical trials with early Alzheimer’s patients, administration of anti-amyloid antibodies reduced amyloid deposits, suggesting that immunotherapies may be promising disease-modifying interventions against Alzheimer’s disease (AD). Specific forms of amyloid beta (Aβ) peptides, for example post-translationally modified Aβ peptides with a pyroglutamate at the N-terminus (pGlu3, pE3), are attractive antibody targets, due to pGlu3-Aβ’s neo-epitope character and its propensity to form neurotoxic oligomeric aggregates. We have generated a novel anti-pGlu3-Aβ antibody, PBD-C06, which is based on a murine precursor antibody that binds with high specificity to pGlu3-Aβ monomers, oligomers and fibrils, including mixed aggregates of unmodified Aβ and pGlu3-Aβ peptides. PBD-C06 was generated by first grafting the murine antigen binding sequences onto suitable human variable light and heavy chains. Subsequently, the humanized antibody was de-immunized and site-specific mutations were introduced to restore original target binding, to eliminate complement activation and to improve protein stability. PBD-C06 binds with the same specificity and avidity as its murine precursor antibody and elimination of C1q binding did not compromise Fcγ-receptor binding or in vitro phagocytosis. Thus, PBD-C06 was specifically designed to target neurotoxic aggregates and to avoid complement-mediated inflammatory responses, in order to lower the risk for vasogenic edemas in the clinic.

Soluble aggregates of Aβ peptides form complex mixtures during the pathology of AD that differ in size, conformation and post-translational modifications 46 . We sought to measure the binding strength of murine PBD-C06 to various mixtures of pE3-Aβ  with Aβ(1-42) oligomers and fibrils, thus attempting to simulate murine PBD-C06 target binding in vivo. When equal pE3-Aβ  and Aβ  fractions were used (Table 3), the binding strength of murine PBD-C06 to these oligomeric and fibrillar mixtures did not change significantly compared to pure pE3-Aβ  aggregates. Even mixed oligomers and fibril preparations containing 5% of pE3-Aβ  were bound with low picomolar avidity, whereas oligomers consisting of 100% Aβ  interact only with nanomolar binding strength. Thus, murine PBD-C06 binds to different aggregates containing pE3-Aβ peptides with high efficiency, a prerequisite for the clearance of these aggregates in vivo.
Humanization and framework substitution. The complementary determining regions (CDRs) within the murine VH and VL sequences were identified by standard antibody structure modeling [47][48][49][50][51][52] . A protein BLAST search 53 was used to identify closely related human VH and VL framework sequences. Based on 82% sequence homology, the human variable kappa sequence BAC01730.1 (Accession Number AB064102.1, family IGKV2-30*01) was selected as a suitable VL framework. For the human VH sequence, AAS85817.1 (accession number AY392875.1, family IGHV1-3*01) was chosen, having a 62% homology to the murine sequence. Subsequently, the murine CDR were grafted onto the human VH and VL genes by standard cloning techniques, thereby  replacing the human CDRs. Finally, human IgG1 heavy chain constant gene and human kappa light chain constant region were fused to the framework genes creating a humanized IgG1:kappa version of PBD-C06, in which the murine CDR sequences are preserved. Initial pE3-Aβ(3-18) binding measurements revealed a 6.5-fold reduced affinity by the humanized PBD-C06 antibody compared to its murine precursor (K D = 6.7 nM vs 43.5 nM see Fig. 2). Prior structural analyses 45 , suggested that framework residues adjacent to the murine CDRs might participate in target binding. Two residues, L41 of the VL and T97 of the VH murine framework, are substituted by F41 and A97 in the humanized framework sequences, respectively (Fig. 3). Since these residues potentially participate in target binding, substitutions of the humanized residues with the original murine residues were studied. Reverting F41 within the human VL framework to the murine equivalent L41 did not improve target affinity (Fig. 2). However, substituting A97 with T97 improved the antigen affinity to 5.3 nM, thereby rescuing the humanized PBD-C06 binding strength to affinities seen with the original murine antibody.
De-immunization. De-immunization was achieved by identifying overlapping peptides of at least 8 amino acids that bind to MHC Class II receptors with high binding strength by standard in-silico methods. Subsequently, critical non-germ-line alleles within each peptide were reengineered to reduce binding affinity. As shown in Fig. 3, one amino acid of the VL sequence at position 53 was identified and reengineered with a conservative substitution (I to T). The mutated VL chain was examined in whole antibody expression and target binding studies and found to have equivalent functionality to that of the original sequence (data not shown).
Within the VH chain sequence seven amino acids were identified that required substitutions. Three of these changes were within the CDR regions and four within the framework region (Fig. 3). Several amino acid substitutions at each indicated VH sequence position were first tested for protein expression individually in combination with the mutated VL chain. Once suitable substitutions were identified, combinations of all VH sequence mutations were tested for protein expression together with the mutated VL chain until an antibody with the VH and VL chain sequences depicted in Fig. 3 was generated. Binding of murine PBD-C06 to Aβ monomers. Sensograms of human pE3-Aβ(3-40) monomers binding to murine PBD-C06 were obtained with a Biacore 3000 SPR instrument. Binding was determined immediately after monomer preparation, using a flow rate of 30 µl/min and injecting 240 µl of 1 nM (light magenta) to 100 nM (green) of pE3-Aβ . The sensograms were recorded for 2500 sec and association and dissociation rates were obtained using the 1:1 Langmuir binding model. Resp.Diff. = Difference in Response Units.   A summary of all allelic substitutions within the VH and VL chain is presented in Fig. 4. Based on our in-silico binding analyses, the immunogenic hotspots within the VL and VH chains could be eliminated except for position 65 of the VH chain. A lysine (K) to Glutamine (Q) substitution at this position was the only choice to reduce the HLA binding strength without sacrificing substantial antigen binding affinity. This change converted the peptide to a germline sequence that should not be recognized due to central tolerance mechanisms.  www.nature.com/scientificreports www.nature.com/scientificreports/ PBD-C06 expression. Upon comparison of the primary amino acid sequence against the VH regions of several antibodies with high or low expression levels, it was noted that certain residue motifs within the VH chain might influence protein stability, expression and folding of the mature antibody. For these purposes four sequence motifs were identified that may negatively influence protein production. Individual mutations and combinations thereof were introduced into antibodies containing all immunogenicity modifications and tested for protein expression in transient HEK293 cell transfections. Exchanging residues at position 12 to 14 from KKS to VKP, position 55 from N to D and position 101 from K to Q of the VH chain, respectively, resulted in expression titers several-fold higher than those obtained with the parental antibodies. Exchanging F to V at position 64 or combining changes at position 12 to 14 (KKS to VKP) with K to Q at position 101 had a negative impact on expression. The combination of the three positive mutations still resulted in improved, but not optimal expression. However this combination greatly reduced antigen binding. Thus, the exchanges at position 12 to 14 and 55 of the VH chain ( Fig. 3) resulted in the highest antibody titers, while retaining full antigen binding and were included into the primary amino acid sequence of PBD-C06, to guarantee optimal expression titers of PBD-C06 during the production phase.
Target binding of the final PBD-C06 candidate. Based on the current results, a final antibody sequence was assembled, which contained all described modifications. This antibody, PBD-C06, is a humanized, de-immunized version of the original hybridoma clone with additional sequence modification to improve protein expression, stability and to inhibit complement activation. PBD-C06 was tested in target binding studies under the same conditions as had been done with the murine antibody ( Table 1). As shown in Fig. 5, the binding strength of PBD-C06 to monomeric pE3-Aβ (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18) is nearly identical to the murine version, as is the high specificity for the pyroglutamated Aβ(3-x) peptides. Based on these binding measurements, we assume that the amino acid modifications in the final PBD-C06 antibody did not change the target binding profile as seen with the murine precursor antibody.
Inactivation of C1q binding. The ADCC effector function of IgG1 antibodies is involved in the phagocytotic removal of insoluble Aβ deposits, while C1q binding to the Fc domain could potentially induce neuroinflammatory processes and vasogenic edema when anti-Aβ antibodies bind to vascular amyloid 54 . In order to minimize the risk for vasogenic edema, we modified the C1q binding site within the Fc region of PBD-C06, while preserving the Fcγ-receptor binding capability. Prior work by Idusogie et al. 55 suggested that a specific alanine substitution at position 322 (K to A) of human IgG1 antibodies significantly inhibited C1q binding and complement activation, while preserving ADCC functionality. C1q binding and Fcγ-receptor binding were measured with a K322A-modified PBD-C06 antibody (Fig. 6). The K322A mutation completely abolished C1q binding in vitro, while preserving binding to the three Fcγ-receptors CD16A, CD32A and CD64.
We next performed cellular antibody uptake experiments to ensure preservation of Fcγ-receptor mediated phagocytosis. The antibody function was assessed by means of the uptake of fluorescent beads by human THP-1 cells coated with pE3-Aβ antigen. The experiments were conducted in the presence or absence of antibody 56 for identification of the opsonisation effect achieved by antibody binding to the antigen coated particles. As shown in Fig. 7, no significant differences in antibody mediated uptake into THP-1 cells could be observed between PBD-C06 antibodies with (K322A) or without (wt) the complement mutation. It can be concluded, that phagocytotic effector function of PBD-C06 is not impaired by the presence of the complement inactivating mutation K322A.

pGlu3-Aβ staining by PBD-C06 variants.
In a further experiment, we assessed PBD-C06's staining specificity of pGlu3-Aβ present in amyloid deposits in the parietal cortex of human AD brain. The development candidate PBD-C06 harboring the mutation to inactivate complement demonstrated a comparable staining pattern to the wild-type antibody (Fig. 8). Thus, the humanization, de-immunization and complement inactivation modifications did not alter the in-situ target binding specificity. www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
In the last years, soluble aggregates of Aβ have been suggested to be the most neurotoxic Aβ species 11 , since the distribution of soluble Aβ aggregates is linked to AD pathology 11,28,[57][58][59][60][61][62][63] . Some in vitro studies point to small Aβ oligomers as the most pathogenic form 63 , while others have linked neurotoxicity to larger oligomers, including protofibrils 64 . However, the soluble Aβ aggregates in the AD brain are likely a polydisperse mixture of monomers of different lengths and post-translational modifications, oligomers of various sizes and conformations, as well as larger than 70 kDa aggregates. Therefore, we have focused on generating an antibody against a neoepitope in the form of the post-translationally modified pyroglutamate ring at the N-terminus (pGlu3-Aβ) of Aβ peptides. Numerous studies point towards a crucial role of pGlu-3 Aβ in AD pathology: (i) It has been shown, that the N-termnal modification leads to rapid formation of small oligomers, which exert enhanced neurotoxicity 34 . This effect appears to have a structural basis, as mixed aggregates consisting of pGlu3-Aβ and Aβ1-42 exert also higher neurotoxicity  www.nature.com/scientificreports www.nature.com/scientificreports/ compared to oligomers consisting only of Aβ1-42. Recent evidence suggests that this effect might be driven by different receptor interaction pattern of these oligomeric species 65 . Moreover, histopathological evidence supports a disease-driving role of pGlu3-Aβ, as the presence of the modification has been shown correlate with cognitive status of the patients and the peptide is strongly accumulating during progression of AD [27][28][29] . Hence, targeting of pGlu3-modified Aβ represents an approach, which is distinguished from other anti-Aβ therapies.  www.nature.com/scientificreports www.nature.com/scientificreports/ We here describe the development of a novel anti-pE3-Aβ therapeutic antibody, PBD-C06, intended for clinical use and treatment of AD patients. Based on crystal structure analyses, the antibody acts as a single epitope, N-terminal-specific antibody against pE3-Aβ 45 and does not bind significantly to unmodified Aβ peptides. Thereby, the physiological functions of Aβ peptides in regulating synaptic activity or acting as an antimicrobial agent 66,67 are preserved. The antibody binds with high specificity and avidity to pE3-Aβ oligomers and fibrils, as well as to mixed aggregates of unmodified Aβ and pE3-Aβ peptides. Aggregates with a low pE3-Aβ content are likely captured by PBD-C06 via protrusion of the N-terminal pyroglutamate ring out of these aggregates 68 . Hence, PBD-C06 has the potential to clear neurodegenerative Ab aggregates believed to be responsible for the AD pathogenesis and cognitive decline in AD patients.
The decision to select PBD-C06 as a clinical candidate was supported by a number of preclinical studies in mouse models with AD-like pathology. The murine precursor antibody inhibited pE3-Aβ(3-42) fibrillation and pE3-Aβ(3-42) oligomer-induced cell death in vitro 45 . The murine IgG1 version of PBD-C06 significantly reduced cerebral plaques in the absence of microhemorrhage and attenuates cognitive impairment in vivo 23,24 . In order to assess the influence of the effector function on efficacy, we compared the murine IgG2a and IgG1 subtypes in a therapeutic trial in APPswe/PS1dE9 mice 69 . We observed a stronger reduction of amyloid deposits and better cognitive preservation with the murine IgG2a antibody than the murine IgG1 antibody, suggesting that the effector function is associated with the therapeutic function. Therefore, we have chosen a human IgG1 isotype sequence, which is analogous to murine IgG2a, as a template for the humanization process.
We noticed that the humanization procedure resulted in a slight loss of target binding strength. The T97 side chain adjacent to the VH CDR3 region and the N-terminal carbonyl oxygen group of pE3-Aβ(3-12) are operating as hydrogen bond donor and acceptor, respectively, and are in close proximity (2.62 Å) to each other (Fig. 2). Hydrogen bond formations between these two participate in the VH interactions of the CDR region with pE3-Aβ (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18). However, these polar interactions are absent in the presence of an alanine and its methyl side chain as present in the humanized sequence. Thus, by reverting alanine back to threonine at position 97, we reconstituted the molecular side chain interactions involved in high binding affinity.
Biological therapeutics require de-immunization in order to prevent the induction of immune response in patients. In the clinic, this immunogenicity induces high levels of anti-drug antibodies (ADA) including neutralizing antibodies 70 . For example, nearly all patients treated with the anti-pE3-Aβ antibody donanemab (LY3002813) develop ADA resulting in accelerated clearance of this antibody 71 . Seven immunogenic hotspots were identified in the humanized PBD-C06 sequence and residue substitutions were made to lower the risk of immunogenicity, while preserving protein stability and expression.
Although the murine antigen binding sequences are not fully conserved in PBD-C06, an identical pattern of antigen binding was found between PBD-C06 and its precursor (Fig. 5). Thus, despite our engineered modifications, we believe that PBD-C06 will potentially clear neurodegenerative Aβ aggregates believed to be responsible for the AD pathogenesis and cognitive decline in AD patients.
In order to address the safety concerns related to vasogenic edemas, human IgG4 or IgG2 backbones or specific Fc mutations have been chosen for antibodies such as crenezumab (Genentech), KHK6640 (Kyowa Kirin), MEDI1814 (Medimmune), ACU-193 (Acumen), ponezumab (Pfizer) or SAR228810 (Sanofi) with the intention to prevent Fcγ-receptor-mediated microglia activation and pro-inflammatory cytokine release, while binding to fibrillar Aβ deposits near cerebral vessels 18 . ARIA-E has not been observed with these antibodies, however, crenezumab and ponezumab failed in slowing cognitive decline 72,73 , suggesting that Fc-effector function might be required for treatment benefits. We have chosen the human IgG1 isotype and eliminated C1q binding by introducing a K322A substitution in the Fc region, while preserving Fcγ-receptor binding. Thus, by inhibiting C1q binding, initiation of the classical complement cascade and microglia-mediated inflammation, which may lead to vascular lesions, is inhibited 74 .
Other IgG1 antibodies which recognize conformational or unique epitopes within the Aβ peptide are aducanumab (Biogen Idec), donanemab (Eli Lilly), Ban2401 (Biogen, Eisai) and gantenerumab (Hoffman-La Roche). Most of these antibodies preferentially target larger soluble Aβ aggregates, although aducanumab also binds to insoluble Aβ fibrils. Aducanumab cleared amyloid plaques in a dose dependent manner 75 , although at the expense of inducing ARIA-E even at low doses. Donanemab and PBD-C06 are IgG1 antibodies against a neo-epitope (pGlu3-Aβ). These antibodies may have a diminished risk for acute inflammatory responses, due to the relatively low abundance of pGlu3-Aβ in amyloid aggregates and due to pGlu3-Aβ's sequential deposition in CAA and plaques 46 . pGlu3-Aβ peptides have been described to be absent in early stages of CAA lesions 76,77 , suggesting that acute CAA-related hemorrhages may be particularly induced by anti-Aβ immunotherapy. Indeed, incidences of ARIA-E have not been reported in Phase I clinical trials with donanemab 71 . In a later update, however, ARIA-E was observed with donanemab administrations 78 . Thus, although targeting pGlu3-Aβ may in principle avoid ARIA, we decided to further reduce the risk of neuroinflammation by inhibiting C1q binding and CR3 binding to macroglia and hence, complement activation and immune cell chemotaxis.
During the submission of this manuscript, Biogen and Eisai first announced the discontinuation of two Phase III aducanumab trials (ENGAGE and EMERGE) for futility 21 . However, a later analysis revealed that the 10 mg/ kg dose of aducanumab did nevertheless cause a significant reduction in decline in clinical dementia outcomes in patients of the EMERGE trial 79 . Similarly, the anti-Aβ antibody BAN2401 (Esai/Bioartic) has progressed into Phase III based on its amyloid plaque lowering and cognitive benefit profile. Although these clinical results are not discussed without reservation, they support the anti-amyloid therapy in general and ask for continued development of immunotherapeutic modalities against AD. We believe that by targeting a neo-epitope (pGlu-Aβ) and by circumventing inflammatory issues (complement inactivation) and immunogenicity (de-immunization), PBD-C06 has great potential to clear the most toxic Aβ aggregates and improve cognition in AD patients at effective doses and with acceptable safety profiles.

Materials and Methods
Cloning, transfection, purification and humanization. The nucleotide sequences of the humanized light chain (LC) and heavy chain (HC) were cloned into the mammalian expression vectors pcDNA3.1 and pOp-tiVEC-TOPO (Invitrogen), respectively. Freestyle ™ CHO-S cells (Thermo Fisher Scientific, Schwerte, Germany) were transfected to transiently express the humanized antibodies according to supplier's instructions. Murine versions of PBD-C06 with and without complement-dependent cytotoxicity (CDC) mutations, as well as experimental forms of de-immunized, humanized PBD-C06 antibodies were expressed transiently in HEK293 cells after transient transfection using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer's instructions. Antibodies were purified on a protein A agarose column (Genscript) after first removing cells by centrifugation and then filtering through a 0.22 µm filter. After loading the supernatants at a rate of 1 ml/min the column was washed with 10 volumes of PBS and 5 volumes of 50 mM Na-acetate, pH 5.5, and then eluted with 50 mM Na-acetate, pH 3.0. Antibody containing fractions were pooled, diluted and diafiltered into PBS using Amicon centrifugal filters (Millipore-Sigma, Burlington, MA).
De-immunization. A computational algorithm 80 was used to identify potential T-cell epitopes within the VH and VL sequences of humanized PBD-C06. Potential T-cell epitopes were predicted by scoring overlapping 9 amino acids sequences within VH and VL against the binding database of 50 HLA-DR alleles. The binding strength of each 9-mer peptide against each of the HLA-DR 50 alleles was determined and a positive binding score assigned to those peptides that displayed a binding strength above an experimentally determined threshold 81 . Subsequently, potential T-cell epitopes were removed by exchanging amino acid residues at critical positions with functional alternatives resulting in a lower binding score or conversion to a germline sequence. All conservative substitutions within the VH and VL chains were tested individually or in combination for protein expression and target binding (data not shown). Generation of mixed Aβ fibrils. Aβ(1-42) and pE3-Aβ(3-42) peptides were dissolved separately in 1,1,1,3,3,3-Hexafluoro-2-propanol (HFIP) to a final concentration of 1 mM according to weighted sample and incubated over-night at RT to generate disaggregated monomers ("seedless" peptides). HFIP was evaporated over-night at RT and each remaining peptide film was dissolved in 100 mM NaOH at a concentration of 500 µM. www.nature.com/scientificreports www.nature.com/scientificreports/ Generation of mixed Aβ oligomers. The generation of mixed oligomers consisting of pE3-Aβ  and Aβ  in different ratios was performed according to a previously described method 82 . Each peptide solution was mixed to a final concentration of 2500 µM oligomers as follows: 100% pE3-Aβ(3-42) = 2500 µM pE3-Aβ ; 50% pE3-Aβ(3-42) = 1250 µM pE3-Aβ(3-42) + 1250 µM Aβ(1-42); 5% pE3-Aβ(3-42) = 125 µM pE3-Aβ(3-42) + 2375 µM Aβ(1-42); 0% pE3-Aβ(3-42) = 2500 µM Aβ(1-42) in a total volume of 15 µl. Three 5 ul aliquot from each oligomer mixture were added sequentially to separate tubes containing 735 µl Ham's F-12 medium (Thermo Fisher Scientific) at RT and immediately vortexed before the next 5 µl aliquot was added. The final four mixed Aβ oligomer preparations were incubated at 4 °C for 18 hours and centrifuged at 13,000 rpm for 10 min at 4 °C. The soluble oligomers present in the supernatants were aliquoted into low binding microtubes (Sarstedt, Germany) and stored at −80 °C until use in Biacore binding experiments. The determination of the binding constants was performed with BIA evaluation software 4.1.1 (GE Healthcare) by simultaneously fitting the association and dissociation phases over all recorded sensorgrams using the "1:1 Langmuir binding" model. In case of baseline drifting, the "1:1 Langmuir binding with baseline drifting" model was used.

SPR (Biacore) antibody binding analyses.
For measuring the binding of murine and humanized PBD-C06 antibodies to multimeric and mixed Aβ aggregates, pure pE3-Aβ oligomers and fibrils were captured with pE3-Aβ-specific antibodies. Pure Aβ  aggregates and Aβ(1-42)-containing mixtures were captured with an N-terminal non-pE3-Aβ-specific antibody. Their densities differed depending on the aggregate: ~800 RU for capturing pE3-Aβ(3-42) oligomers, ~15,000 to 19,000 RU for capturing mixed oligomers and ~1,100 RU for capturing mixed fibrils. The sensorgrams were evaluated from the first injection showing an antibody binding onwards using a Single Cycle Kinetics model.
Immunohistochemistry. Human Alzheimer's disease brain tissues were collected postmortem at the time of autopsy, having obtained informed consent for study participation from the next of kin, and following protocols approved by the Partners Human Research Committee at Brigham and Women's Hospital (BWH, Boston, MA). We confirm that all methods were performed in accordance with the relevant guidelines and regulations.
Human AD brain blocks were collected at autopsy and fixed in 10% neutral buffered formalin for 1-2 hours (brief fixation). Tissue blocks were processed for paraffin embedding and then sectioned at 10 μm for IHC analysis at BWH. Sections were deparaffinized in two changes of Histo-Clear (National Diagnostics, Atlanta, GA) and rehydrated in graded ethanol solutions. Endogenous peroxidase activity was quenched with 0.3% H 2 O 2 in methanol for 10 minutes. All paraffin sections were pretreated with 88% formic acid for 5 minutes for antigen retrieval. Sections were subsequently washed with water, and incubated with 10% goat serum for 20 minutes at room temperature and primary antibody (0.3 μg/ml) overnight at 4 °C. After incubation for 30 minutes at room temperature with a biotinylated secondary antibody (Vector Laboratories, Burlingame, CA), immunoreactivity was visualized with the VectorElite horseradish peroxidase ABC kit (Vector Laboratories) and DAB (Sigma-Aldrich) as chromogen.

Data availability
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.