Generation of a stable, disulphide-free, autonomous antibody variable domain for intracellular expression.
The 4D5 VH domain was chosen as the parental scaffold for stabilization, as its parental IgG trastuzumab is an approved FDA drug with a favourable immunogenic and stability profile.6 The 4D5 VH domain was fused to an N-terminal signal sequence for periplasmic expression in E. coli and a random mutagenesis library of the VH ORF was created. This was then used to initiate 2 rounds of selection using Hot-CoFi11 (Fig. 1A and 1B), resulting in the isolation of 53 stabilized VH domain variants. Among these, VH-36 and VH-36 were the most thermally stable, with TCAGGs of 73.4°C and 72.9°C, respectively, an increase of over 20°C in comparison to the parent WT clone (Fig. 1C). The thermostable VH-36 clone was then selected as the starting template for the evolution of a disulphide-free VH domain suitable for intracellular cell expression. Three residues in VH-36, C22 and C92, as well as the core residue A24, were randomised with all 20 amino acids to generate a library suitable for screening. The signalling peptide, stII, was also removed from the VH-36 expression construct (termed VH-36i) to ensure expression in the cytoplasm and allow screening for autonomously stable VH domains in a cellular reducing environment. This library was screened using the Hot-CoFi methodology and five unique variants of VH-36i were identified (Fig. 1A and 1C). These were then expressed and purified, and their thermal point of denaturation (Tm) determined by differential scanning fluorimetry (DSF), with VH36i.1 demonstrating the highest thermal stability with a Tm of 58°C. VH36i.1 possessed the mutations C22S, A24C and C92T (Fig. 1A and 1C).
Optimization of a disulphide-free autonomous antibody variable domain for phage display.
The VH-36i.1 clone and its parent templates (VH-36 and VH-36i) all contained the mutation A100bP within the CDR-H3 loop. An undesirable feature with respect to target binding, as it restricts the conformational space open to randomized CDR-H3 loop regions and in turn decreases their probability of interacting successfully with target molecules. Therefore, the CDR-H3 region from position W95 to P100b was removed and replaced by the sequence SSSA, creating the new construct VH-37i, which had significantly reduced thermal stability as determined by DSF (Tm=46°C). VH-37i, was then used to create a random mutagenesis library for CoFi screening10 to compensate for this loss of protein stability. From this process 4 variants were identified with VH-37i.1 and VH-37.2 exhibiting the highest thermal stabilities (Tm), approximately 55.7°C for both (Fig. 1A and 1C). Both clones contained a single mutation with VH-37i.1 bearing a A78V mutation in the core of the VH domain, whilst VH-37.2 contained the mutation G93V located near the CDR-H3 loop (Fig. 1D). Both mutations were combined to generate VH38i with a Tm of approximately 57.3°C. To remove the final cysteine in VH-38i at position 24 it was either mutated to the Ala residue found in the WT template, Ile, Val, Tyr, or Trp. The VH-38i variants containing either C24I or C24L substitutions were the most stable with improved Tms of 58.3°C and 58.3°C, respectively compared to the VH-38i template molecule (Fig. 1C).
VH Domain Phage Display library construction and Delineation of the VH-1A2 and VH-1C5 interaction sites on eIF4E
A library of VH domains were then displayed on the pIII protein of M13 phage, where the CDR1 and CDR2 of the VH-38i template were randomized conservatively, whilst CDR3 was randomized with sequences of different lengths and biased towards residues serine and tyrosine with approximately 20% frequency each. Phage display selection against purified human full-length eIF4E led to the identification of VH Domains VH-1A2 and VH-1C5 as the highest affinity binders with Kds of 115.2 ± 4.4 nM and 154.3 ± 70.3nM, respectively (Fig. 2A, Table 1). The phage selected CDR3 loop sequences of the VH-1A2 and VH-1C5 domains shared little similarity to the interacting motifs of known eIF4E binding proteins such as eIF4G1 and the 4E-BP family (YXXXXLΦ, termed the canonical motif) (Fig. 2B). Competitive based fluorescence anisotropy experiments were performed that mapped the binding of VH-1A2 and VH-1C5 to the eIF4G interaction site (Fig. 2B).
Table 1
Binding and kinetics parameters of various VH domains against eIF4E. Kd, kon and koff values were derived from SPR experiments using either single cycle or multi cycle injection experiments against eIF4E amine coupled to a CM5 sensorchip. Parameters were derived using a 1:1 site binding model. However, with regards to 4E-BP14ALA, binding parameters were initially determined using a 1:1 site binding model, which generated a poor fit and physically irrelevant Kon and Koff values beyond the detection limit of the machine. This analysis was superseded with a 2-state analysis, which generated more meaningful kinetic parameters. The use of a 2-state model is supported by following reasons: 1) 4E-BP14ALA undergoes a disorder to order transition upon binding eIF4E and, 2) it interacts at eIF4E through 2 binding sites.
|
Kd (nM)
|
kon (S− 1M− 1)
|
koff(S− 1)
|
VH-1A2
|
115.2 ± 4.4
|
3.43 x 105 ± 3.75 x 104
|
3.78 x 10− 2 ± 3.17 x 10− 3
|
VH-1C5
|
154.3 ± 70.3
|
2.85 x 105 ± 8.5 x 104
|
4.11 x 10− 2 ± 6.99 x 10− 3
|
VH-1C5D104A (VH-M4)
|
4.8 ± 1.7
|
2.46x 106 ± 3.89 x 104
|
1.18 x 10− 2 ± 3.90 x 10− 3
|
VH-1C5D104A/S108R
|
0.94 ± 0.04
|
2.63 x 106 ± 2.98 x 105
|
2.48 x 10− 3 ± 3.79 x 10− 4
|
VH-1C5D104A/F120I (VH-S2)
|
0.64 ± 0.06
|
1.18 x 106 ± 1.85 x 105
|
7.64 x 10− 4 ± 1.80 x 10− 4
|
VH1-C5D104A/S108R/F120I (VH-S4)
|
0.057 ± 0.004
|
5.18 x 106 ± 8.55 x 105
|
2.96 x 10− 4 ± 6.47 x 10− 5
|
4E-BP14ALA (1:1)
|
0.60 ± 0.14
|
1.05 x 1010 ± 1.48 x 109
|
5.93 ± 0.56
|
|
|
kon1
|
kon2
|
koff1
|
koff2
|
4E-BP14ALA (2 state)
|
0.68 ± 0.13
|
8.35 x 106
± 4.90 x 106
|
1.70 x 10− 2
± 1.08 x 10− 2
|
3.19 x 10− 4
± 7.91 x 10− 5
|
1.84 x 10− 4
± 7.00 x 10− 5
|
Structural Characterization Of Vh-1c5 Interaction Site With Eif4e
Crystals were obtained for eIF4E complexed with m7GTP and VH-1C5 (Figure S1), which revealed that VH-1C5 bound eIF4E at a position that overlaps with the eIF4G interaction site. The phage selected CDR3 loop (residues 111–120) of VH-1C5 forms a highly folded small domain structure that interacts directly with eIF4E, whose structure contrasts sharply with the ‘L’-shaped conformers formed by 4E-BP1 and eIF4G derived linear peptides when in complex with eIF4E (Fig. 2C and 2D)16. The random coil section (residues 101–107) of the CDR3 loop orientates the VH domain residue L105 into a position that mimics the interactions made by the conserved hydrophobic residue in the eIF4E interaction (residues M60 and L630 in 4E-BP1 and eIF4G1, respectively) (Fig. 2E). The random coil section of the VH domain also forms additional contacts with a hydrophobic region located on eIF4E unexploited by the canonical peptides17,18 e.g., the side chain of A102 forms a hydrophobic contact with W73, whilst the sidechain of D104 packs against Y76 and L131. Y117, located on the helical segment (residues 108–117) of the CDR3 loop, forms hydrophobic packing interactions with L135 and a sidechain to sidechain hydrogen bond with E132 on the surface of eIF4E (Fig. 2E). These interactions replace the conserved electrostatic interaction mediated by R55 and R625 of the 4E-BP1 and eIF4G1 peptides with E132, respectively. F220, positioned on the short helical turn motif (residues 119–121) that precedes the loop re-joining the main body of the VH1C5 fold, forms several hydrophobic contacts with the eIF4E residues L39, V69 and I138 and a backbone hydrogen bond interaction with the indole side chain of W73 on eIF4E (Fig. 2F). These interactions are replicated by the conserved Y and L residues of the eIF4E interaction motif (Y54/624 and L59/629 of 4E-BP1 and eIF4G1, respectively) with the backbone carbonyl of the L forming a hydrogen bond with W73.
Alanine Scanning Mutagenesis Reveals V H 1C5 Residues Distal to the Interaction Interface are Critical for Binding
Alanine mutagenesis scanning of the CDR3 library insert (Fig. 3) was performed using yeast display. These experiments revealed two classes of mutants that adversely affected the binding of VH1C5 to eIF4E (Fig. 3A and 3B). The first class of mutations describe residues identified in the crystal structure that are in direct contact with eIF4E (K102, A103, D104, L105, T116, Y117 and F120). These results showed high correlation with residues that were identified through computational molecular dynamics simulations to significantly contribute energetically to the eIF4E:VH-1C5 complex interface (Fig. 3C and Table S1). Interestingly, both alanine mutagenesis scanning and binding decomposition analysis identified that D104 forms unfavourable interactions at the eIF4E and VH-1C5 interface, and when replaced with alanine resulted in a VH-1C5 variant (VH-M4) that had an approximate 17-fold improvement in its affinity towards eif4E (Fig. 3D, Table 1) with a Kd of 4.8 ± 1.7 nM. Inspection of the eIF4E: VH-1C5 structure revealed that the residues in proximity to D104 are primarily hydrophobic and that the D104A mutation allows for more favourable van der Waals interactions to occur.
The second class of mutations describe the alanine scanning replacement of residues V101, P110, V113, V114 and F122 that abolished eIF4E binding but were not located at the VH-1C5 interface. These residues form a distinct hydrophobic cluster that interact with several hydrophobic residues located on the β-sheet face of the immunoglobulin fold of the VH domain (V39, L47 and W49) (Fig. 3E). The sensitivity of VH-1C5 binding to eIF4E when this cluster is mutated suggests that these residues play a critical role in ensuring the correct folding of the CDR3 loop and spatial orientation of the contact residues. It also implies that the CDR3 loop forms a highly stable fold, a conclusion further supported by MD simulations that shows that the CDR3 loop structure of the unbound form deviates negligibly from the bound form (Fig. 3F). A salt bridge also forms between residues R52 and D37 at the interface between the CDR3 loop and the VH domain fold, which orientates the R52 to form long range electrostatic interactions with the backbone carbonyls of S107 and S108 further rigidifying the CDR3 loop.
Evolution of an Ultra High Affinity Mini-Protein Inhibitor of the eIF4E:4G interaction
The improved VH-M4 binder (Kd = 4.8 ± 1.7 nM) was used as a template sequence to generate a randomly mutagenized library for affinity maturation to identify improved binders against eIF4E using yeast display (Fig. 4A). Three rounds of kinetic selection were performed with increasing incubations times of 8 mins, 60 mins and 110 mins to increase the competitive pressure of unlabelled eIF4E against the VH domain library complexed to fluorescent eIF4E (Fig. 4B). The final round of kinetic selection resulted in the identification of only three eIF4E binders: VH-1C5D104A/S108R/F120I, VH-1C5D104A/S24G/F120I and VH-1C5D104A/Y97C/F120I (Fig. 4C). The VH-1C5D104A/S108R/F120I clone (termed VH-S4) dominated the final round with a frequency > 90% and was selected for scale up and SPR analysis, which revealed that it interacted with a sub-nanomolar Kd of 0.057 ± 0.004 nM (Fig. 4D, Table 1). The 2 substitutions identified in the VH-S4 sequence were then individually introduced into VH-1C5-M4 and their binding to eIF4E assessed. SPR analysis revealed that VH-1C5 D104A/S108R and VH-1C5D104A/F120I (termed VH-S2) possessed Kds of 0.94 ± 0.04 nM and 0.64 ± 0.06 nM, respectively (Fig. 4, Table 1). Preliminary examination and MD simulations of the VH-1C5:eIF4E crystal structure with the identified mutations highlighted no substantial change in the energetics profile of the interaction (Figure S2) but did indicate that the F120I mutation most likely optimised the hydrophobic contacts between both proteins and allowed a more optimal engagement to occur. However, with regards to the S108R mutation, MD simulations showed only a transient interaction with R128 of eIF4E (Table S1).
The crystal structure of eIF4E: VH-1C5D104A/S108R/F120I (VH-S4) complex was resolved, which enabled confirmation that the F120I mutation improved the complementarity of fit between the two proteins (Fig. 4E and S3). The elucidated complex structure also showed that the S108R mutation did not result in a direct interaction with the surface of eIF4E, but rather that it engaged a structured water network that forms an h-bond to the backbone carbonyl of R128 on eIF4E (Fig. 4F and S3). Additionally, S108R formed a cation-π interaction with the sidechain of the neighbouring F112, further helping to stabilise the local fold of the CDR3 recognising eIF4E. The eIF4E: VH-1C5 D104A/S108R/F120I (VH-S4) crystal structure also revealed that the substitution of D104 with alanine results in a significant conformational change in Y76 upon the surface of eIF4E (Fig. 4G and S3). This change is also associated with the subtle migration of several structure waters that enable the local H-bond network around the Y76 residue to be maintained. The D104A mutation removes the unfavourable packing of the negatively charged D104 side chain from against Y76 and allows a more optimal packing arrangement to occur (Fig. 4G and S3).
VH-S4 Disrupts eIF4F Complex Formation and Cap-Dependent Translation In Vitro
VH-S4 was expressed in HEK293 cells and shown to immune-precipitate endogenous eIF4E more strongly than the VH domain mutants with weaker affinities (VH-1C5, VH-M4 and VH-S2) for eIF4E (Fig. 5A). A VH-1C5 negative control (VH-1C5SCRM), where the residues in the CDR3 loop were scrambled, was also included that failed to pull-down eIF4E demonstrating the specificity of the VH domains for eIF4E (Fig. 5A). The ability of VH-S4 to interact with eIF4E was also compared to a constitutively active 4E-BP1 construct (termed 4E-BP14AlA), where the phosphorylation sites (Thr37, Thr46, Ser65, Thr70)19 responsible for modulating its binding with eIF4E were mutated to alanine. The Kd of 4E-BP14ALA against eIF4E was determined to be 0.68 ± 0.13 nM approximately 10-fold weaker than the VH-S4 interaction with eIF4E, with a substantially weaker off-rate (Table 1 and Figure S4). Both constructs exhibited comparable levels of activity in immuno-precipitating eIF4E, whilst the 4E-BP14ALA negative control (termed 4E-BP1YLM) had negligible effects. VH-S4 was then shown to more efficiently disrupt the eIF4G-eIF4E interaction in cells than the weaker affinity VH domain variants (1C5, S2 and M4) using the NanoBit eIF4E:eIF4G606–646 cell-based assay20 (Fig. 5B). 4E-BP14ALA was also assessed in the NanoBIT assay and again was similar in activity to VH-S4. All the VH-domains except the control were then shown to inhibit cap-dependent translation in a bicistronic reporter assay, with the magnitude of cap-independent inhibition again correlating to the affinity of the eIF4E interacting VH domains (Fig. 5C). It is well established that disruption of the eIF4F complex prevents eIF4G mediated phosphorylation of eIF4E by Mnk1.21 Therefore, in parallel, lysates were co-prepared from the cells used in the NanoBIT assay and the levels of phosphorylated eIF4E detected (Fig. 5C). eIF4E phosphorylation as expected closely followed the levels of eIF4F complex disruption measured in the NanoBit eIF4E:eIF4G606–646 system. 4E-BP14ALA was also tested in both the bicistronic and eIF4E phosphorylation assays, where it specifically decreased eIF4F complex formation and phosphorylation levels in a similar manner to the VH-S4 domain (Fig. 5C).
The effects of eIF4F complex disruption directly upon cap-dependent protein translation were monitored by measuring Cyclin D1 protein levels22,23. Increasing amounts of VH-M4 and VH-S4 expression plasmid were transfected into mammalian cells with a concomitant inhibitory effect on eIF4E phosphorylation and cyclin D1 protein levels (Fig. 5D), with the more tightly binding VH-S4 domain (Kd = 60 pM) having a more profound effect. This effect of eIF4F disruption upon cyclin-D1 translation and eIF4E phosphorylation was also reflected by transfection experiments with 4E-BP14ALA (Fig. 5E and 5F). m7GTP mediated pull down experiments, using HEK293 transfected cells, verified that all the VH domains (S4, M4, S2 and 1C5) and 4E-BP14ALA, except for the controls, were able to competitively displace eIF4G from eIF4E and decrease the amount of endogenous eIF4F complex detected (Fig. 5E).
Vh-s4 Modulates Eif4f Mediated Signalling Pathways Specifically
eIF4E inhibition has been shown to reduce the expression of malignancy-related proteins (e.g., cyclin D124, Mcl-125 and BCL-xl26–28). Additionally, the most reported biological consequences of eIF4E inhibition are significant reductions in cellular proliferation and induction of apoptosis. To examine the cellular effects of VH-S4 and 4E-BP14ALA mediated inhibition of eIF4E, stably transfected inducible expression systems were constructed for both proteins in A375 melanoma and MBA-MD-321 breast carcinoma cells. VH-S4 and 4EBP14Ala were both induced with doxycycline for 24 hours in both cell lines and were shown to dramatically decrease eIF4E and eIF4F complex formation (Figure S5A) in comparison to mock control cells. Both proteins when induced for over 7 days also caused significant decreases in the cellular proliferation and viability (Figs. 6A, 6B, S5C and S5D). The most dramatic effects were seen in A375 cells, where both 4E-BP14ALA and VH-S4 decreased cellular proliferation and viability to the same extent (Fig. 6A and 6B). In contrast, VH-S4 was less efficacious than 4E-BP14ALA in MD-MBA-231 cells, despite disrupting eIF4F complex formation to the same extent (Figure S5A).
We next assessed the effects of both proteins on the expression of several proteins (Cyclin-D1, Mcl-1 and Bcl-xl) whose translation is regulated by the eIF4F complex in A375 cells (Fig. 6C). Both VH-S4 and 4E-BP14ALA when induced for 24 hours resulted in significant decreases in the protein expression levels of cyclin D124 and of MCL-125 but had little effect on Bcl-xl 26–28 levels. Similar effects on Mcl-1 and Cyclin D1 levels were observed in the MDA-MB-231 cell line (Figure S5B). Interestingly, A375 cells demonstrated no evidence of apoptosis as measured by PARP cleavage with the induction of either miniprotein, post 24 and 72 hours (Fig. 6C). This result suggests that the down-regulation of Mcl-1 is insufficient to induce apoptosis in A375 cells and that the decrease in cellular proliferation measured with both VH-S4 and 4E-BP14ALA is principally driven by the reduction in Cyclin-D1 protein levels (Fig. 6C). Additionally, both VH-S4 and 4E-BP14ALA reduced the total protein expression of 4E-BP2 with negligible effects on 4E-BP1 levels. Protein expression levels was reassessed post 72 hours doxycycline induction, which revealed that Mcl-1 was no longer significantly repressed but showed that 4E-BP1 levels were reduced with a concomitant decrease in its phosphorylated forms (Figure S6A). This lack of sustained decrease in Mcl-1 levels offers further explanation of the lack of cellular apoptosis with eIF4F complex disruption. In parallel experiments, both VH-domains and 4E-BP14ALA inhibited global protein synthesis by approximately 50% as determined using puromycin pulse chase experiments (Figure S6B). Results that correlated with the significant effects of eIF4F complex disruption upon cellular proliferation (Fig. 6A). We also verified the role of eIF4F in mediating STAT1 levels in IFN-γ treated A375 cells (Fig. 6D)29, where selective induction of VH-S4 or 4E-BP14ALA in IFN-γ treated cells reduced STAT1 protein levels.
The specificity of the VH-S4 domain was assessed by examining its effects on the pathways (AKT/mTORC and RAS/ERK) that regulate the eIF4F complex. Many of the chemical reagents used to study the biological function of the eIF4F complex modulate these pathways, and as a result elicit eIF4F independent effects e.g. PP24230, an ATP competitive inhibitor of mTORC1/2 that leads to dephosphorylation of both its downstream targets, 4E-BP1 and S6 kinase. A375 cells were treated with either PP242 or staurosporine (a non-selective kinase inhibitor)31 and their effects on eIF4E, AKT, ERK and rS6 phosphorylation compared to disruption of the eIF4F complex by both 4E-BP14ALA and VH-S4 (Fig. 6C). Both miniproteins, caused dephosphorylation of eIF4E in a manner similar to PP242 through inhibition of eIF4G mediated MNK1 phosphorylation.20 As expected, PP242 reduced 4E-BP1 phosphorylation to mediate its known effects on eIF4F complex disruption30. However, PP242 unlike the miniproteins also reduced phosphorylation of rS6 and AKT phosphorylation though its effects on mTORC12 (Fig. 6C)30,32. A375 cells treated with 0.2 µM of the broad kinase inhibitor staurosporine resulted only in the dephosphorylation of S6 kinase. Neither the miniproteins nor the small molecules tested affected ERK phosphorylation.
The effects of small molecules upon the eIF4F complex were extended to a wider set of compounds (including the eIF4E inhibitor, 4EGI-1 and the Mnk Kinase inhibitor, CPG-57380) and assessed at 72 hours (Fig. 6E). The effects of PP242 and staurosporine on protein expression levels of Mcl-1 at 24 hours (Fig. 6C) were alleviated by 72 hours (Fig. 6E), whilst maintained with respect to cyclin-D1, 4E-BP1 and 4E-BP2. CPG-57380 as expected resulted in dephosphorylation of eIF4E at 72 hours (Fig. 6E) with negligible effects on cyclin-D1 or 4E-BP1/2 protein levels. Additionally, PP242 and staurosporine induced apoptosis at 72 hours as indicated by PARP cleavage (Fig. 6E and Figure S6C)) in contrast to the miniproteins demonstrating that specific inhibition of the eIF4E:4G interface does not lead to apoptosis and that this is the result of other modes of action by either small molecule. Interestingly 4EGI-1, a molecule that allosterically disrupts binding of eIF4G to eIF4G, also elicits PARP cleavage and promotes eIF4E phosphorylation in contrast to VH-S4 and 4E-BP14ALA.