Selection of antibodies against eqIL5 via antibody phage display
The next step of the development of a therapeutic antibody against eqIL5 was the selection of antibodies that bind to this target via antibody phage display. In this study, three different panning approaches were applied in order to select a large number of diverse binders: Firstly, panning in multititer plates (MTP) with immobilized antigen [25], secondly, panning in solution with biotinylated antigen [26] and thirdly, capture panning with human Fctagged antigen. For all three methods the human naïve antibody gene libraries HAL9 (lambda) and HAL10 (kappa) were used as starting material [27]. Table 1 summarizes the number of individual binders selected by the three panning strategies in relation to the number of clones that was tested in the Screening ELISA.
Table 1
Selection of unique binders against eqIL5 via different antibody phage display techniques.
Library
|
Panning in MTP
|
Panning in solution
|
Capture Panning
|
HAL9
HAL10
|
12/368
3/368
|
15/276
4/184
|
2/92
0/92
|
With all three panning techniques, individual binders could be selected against the target eqIL5. The average hit rate for panning in MTP was 2.0%, for panning in solution 4.1% and for capture panning 1.1%. 29 lambda binders and 7 kappa binders were selected, which results in an overall hit rate of 3.9% for the HAL9 and 1.1% for the HAL10 library. In total, 36 binders were selected against eqIL5. One of these binders (NOL461A1) was selected using all three different techniques thereby reducing the number of unique binders to 34. In a next step, these antibodies were cloned and produced in the human single chain fragment variable (scFv)Fc format for further analysis. Six antibodies could not be produced in this format and were therefore excluded from further analysis. This resulted in 28 remaining scFvhFc antibodies against eqIL5 for a first functional screening.
Selection of lead candidates with a cellular in vitro inhibition assay
In the next step, the binding antibodies were tested regarding their functionality. For this purpose, a cellular inhibition assay was established in order to test whether the binding of the selected antibodies to the antigen inhibits the interaction of the interleukin and its receptor on a cell surface. Expi293F suspension cells were transfected transiently with the eqIL5 receptor DNA subunits IL5RA (Uniprot A0A3Q2L5Z7) and CSF2RB (Uniprot F7DHE0) and co-transfected with eGFP (GenBank AEI54555.1) to differentiate transfected from non-transfected cells. The binding of the interleukin to its receptor on the cells was detected via the 8x His-tag and cells were analyzed by flow cytometry (Supplementary Fig. S2).
Figure 1 presents the relative binding of the interleukin to the cell after pre-incubation with the individual antibody. As a reference for 100% binding, antigen binding to the cell without application of test antibody was measured. The selected antibodies in scFvhFc format were screened in the cellular inhibition assay using 1000 nM antibody and 5 nM antigen (molar ratio 200:1) (Fig. 1a). The cut-off for inhibiting antibodies was set at 20% so that only antibodies with a significant inhibition effect were considered for next steps. 16 out of the 28 antibodies remained below this cut-off. Nine of these antibodies were selected by panning in MTP, five by panning in solution and one by capture panning. Additionally, the antibody NOL461A1 selected with all three panning strategies was one of the best three inhibiting antibodies. To further characterize the inhibition effect of these 16 antibodies, they were tested in concentrations ranging from 1000 nM to 0.32 nM with a constant antigen concentration of 5 nM (related to dimer) (molar ratio 200:1–0.064:1). Here, a comparison was made for the antibodies in the scFvhFc format (Fig. 1b) and the final equine IgG6 (eqIgG6) format (Fig. 1c). For the eqIgG6 format all constant domains of the antibodies were replaced with equine domains (IGHC IMGC000040, IGLC [28], IGKC IMGT000053), only the VH and VL remained of human origin. The antibodies NOL481C10, NOL481E1, NOL1792A8 and NOL185C12 could not be produced as eqIgG6 and were therefore excluded from further selection. In both formats an unrelated isotype control (SH1351C1 anti Ncadherin propeptide) was included.
In order to compare the inhibition effect, the antibody concentration necessary to reduce the relative binding to 50% (IC50) was determined. The tables presented in Fig. 1b and 1c summarize the IC50 values for the selected antibodies in scFvhFc and eqIgG6 format. IC50 values for the scFvhFc antibodies ranged from 3 nM (NOL461E1) to 163 nM (NOL461E4) and for eqIgG6 antibodies from 12 nM (NOL481D5 and NOL1621-F5) to 522 nM (NOL461E4). The IC50 value for NOL1791-H8 as eqIgG6 could not be determined due to its low inhibition effect.
Nearly all tested scFv-hFc antibodies had lower IC50 values compared to the eqIgG6 format. To gain a better understanding of this phenomenon, two antibodies (NOL461A1 and NOL461E4) were compared as scFvhFc, scFveqFc and eqIgG in this assay (Supplementary Fig. S3). Antibodies in the equine scFvFc and human scFvFc format showed a comparable inhibition effect, which was different from the inhibition effect of the eqIgG antibodies. This indicates that the antibody format but not the Fc-part is responsible for the different inhibition behavior.
The selection of lead candidates for further development was based on the inhibition of the antibodies as eqIgG6 since the full IgG is the preferred format for therapeutic application. The two selected eqIgG6 antibodies were NOL481D5 (selected by panning in solution) and NOL1621F5 (selected by panning in MTP), both with an IC50 of 12 nM (molar ratio of antibody (bivalent) to antigen (dimer): 2.4).
Besides binding and inhibition, stability and specificity are also of great importance for an antibody for therapeutic application. Before further experiments were performed, these characteristics needed to be confirmed in order to enhance the chances of obtaining final candidates with the desired features. Thus, the two antibodies NOL481D5 and NOL1621F5 were characterized in the eqIgG6 format regarding their stability in titration ELISA, inhibition assay and size exclusion chromatography (SEC) (Supplementary Fig. S4) and regarding their specificity in an unspecificity ELISA (Supplementary Fig. S5). In summary, only slight decrease of stability was detected when antibodies were stored up to 28 days at 37°C at a concentration of 0.3 mg/mL. Also, no unspecific behavior of the antibodies was observed in the unspecificity ELISA.
In vitro affinity maturation for improvement of lead candidates
Next, an in vitro affinity maturation was performed with the lead candidates in order to improve their binding activity and consequently their inhibition effect. Within the scope of library generation from a single parental sequence many approaches have been developed for the introduction of random mutations at the nucleotide level, e.g. error prone PCR [29], mutator bacterial strains [30] and site-specific mutagenesis of the complementary determining regions [31]. Here, we chose several rounds of error-prone PCR to generate scFv mutation libraries. With these scFv mutation libraries a panning approach with low amount of antigen and extended washing and incubation steps was performed in order to select antibodies with improved binding activity. A screening ELISA with scFv supernatant was used to find clones with the highest ELISA signals. In addition, it was confirmed by Western Blot that the signals were not significantly biased by the producibility of the scFvs. In total, 3.8% of mutants derived from NOL481D5 had increased signals compared to their parental scFv (up to factor 1.5). Regarding the mutants derived from NOL1621F5 there were 6.6% of hits with increased signals (up to factor 2.6). Seven mutants derived from NOL481D5 and four mutants derived from NOL1621F5 were selected, all with a signal increase > 1.3x compared to the corresponding parental scFv. An amino acid sequence alignment of the VH and VL of these affinity-matured antibodies is presented in Fig. 2.
The number of mutations at the amino acid level in the whole scFv region ranged for mutants derived from NOL481D5 from 1 to 5 with an average of 2.7 amino acid exchanges (Fig. 2a). Mutants derived from NOL1621F5 had 2 to 4 amino acid exchanges with an average of 3.5 (Fig. 2b). Overall, the position of the mutations was spread over the whole sequences with no significant accumulation in the CDR regions. Only single positions of amino acid substitutions were found in several affinity-matured antibodies (NOL481D5: VH at position 101, VL at position 21 and 96; NOL1621F5: VH at position 30 and 97).
The antibody mutants were cloned and produced as equine IgG6 antibodies in Expi293F suspension cells for final characterization. The affinity-matured antibodies were analyzed by titration ELISA (Fig. 3a) and cellular inhibition assay (Fig. 3b) including the EC50 and IC50 values in order to compare their binding activity and inhibition effect.
In the titration ELISA all tested antibodies bound specifically to eqIL5 and not to BSA. Overall, the antibodies derived from NOL481D5 had a stronger binding activity with EC50 values between 0.4 nM and 2.1 nM compared to the antibodies derived from NOL1621F5 with EC50 values ranging from 7.1 nM to 24.2 nM. Both parental antibodies belonged to the least affine antibodies compared to their respective mutants, only single mutants showed weaker binding. The affinity maturation improved the EC50 for NOL481D5 from 1 nM to 0.4 nM (factor 2.5) and for NOL1621F5 from 9.0 nM to 7.1 nM (factor 1.3).
In the cellular inhibition assay the IC50 values ranged from 3 nM to 9 nM including all tested antibodies. Most mutants had a reduced IC50 value compared to both parental antibodies. The affinity maturation improved the IC50 value for NOL481D5 from 6 nM to 3 nM (factor 2.0) and for NOL1621F5 from 7 nM to 4 nM (factor 1.8). Mutants derived from the different parental antibodies against eqIL5 were similarly effective in the inhibition assay, despite different binding behavior in the titration ELISA. This could be explained by the assumption that the antibodies recognize different epitopes. In conclusion, the weaker binding in ELISA of mutants derived from NOL1621F5 is no criterion for exclusion since the main aspect for selection of final antibodies is the functionality tested in the cellular inhibition assay.
Many mutants were in a similar range regarding their inhibition effect. Considering the parameters IC50 value, curve progression and producibility of the antibodies, we chose NOL2262D10 (IC50: 4 nM; producibility: ~ 70 mg/L) and NOL2311A3 (IC50: 4 nM; producibility: ~ 30 mg/L) as most promising candidates. Further mutants, such as NOL2252C1 (IC50: 4 nM; producibility: ~ 15 mg/L) and NOL2312E2 (IC50: 3 nM; producibility: ~ 25 mg/L), could be considered as potential backup candidates.
The two antibodies NOL2262D10 and NOL2311A2 were produced in a 250 mL scale in Expi293F suspension cells. Here, the antibody NOL2311A3 was not stable in concentrations > 0.8 mg/mL. The antibody solution showed turbidity indicating that aggregation of the antibody occurred. Also, the antibody concentration measured spectrophotometrically was decreasing over time. Attempts to stabilize NOL2311A3 by changing the pH value were not successful. Therefore, NOL2262D10 was chosen as final candidate for further studies. This antibody provides all required properties for further clinical development: A strong inhibition effect (IC50 = 4 nM), nanomolar binding activity to the target (EC50 = 8.8 nM), satisfactory producibility (~ 70 mg/L in transient expression system) and high stability (SEC data of a stability test in Supplementary Fig. S6).