Emma E.
Cawood‡
ab,
Emily
Baker
cd,
Thomas A.
Edwards
aei,
Derek N.
Woolfson
*bcf,
Theodoros K.
Karamanos
*g and
Andrew J.
Wilson
*abh
aAstbury Centre for Structural Molecular Biology, University of Leeds, Woodhouse Lane, Leeds LS2 9JT, UK. E-mail: a.j.wilson.1@bham.ac.uk
bSchool of Chemistry, University of Leeds, Woodhouse Lane, Leeds LS2 9JT, UK
cSchool of Biochemistry, University of Bristol, Medical Sciences Building, University Walk, Bristol BS8 1TD, UK. E-mail: d.n.woolfson@bristol.ac.uk
dBrisSynBio, University of Bristol, Life Sciences Building, Tyndall Avenue, Bristol BS8 1TQ, UK
eSchool of Molecular and Cellular Biology, University of Leeds, Woodhouse Lane, Leeds LS2 9JT, UK
fSchool of Chemistry, University of Bristol, Cantock's Close, Bristol BS8 1TS, UK
gDepartment of Life Sciences, Imperial College London, London, SW7 2BX, UK. E-mail: t.karamanos@imperial.ac.uk
hSchool of Chemistry, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK
iCollege of Biomedical Sciences, Larkin University, 18301 N Miami Ave #1, Miami, FL 33169, USA
First published on 6th June 2024
A significant challenge in chemical biology is to understand and modulate protein–protein interactions (PPIs). Given that many PPIs involve a folded protein domain and a peptide sequence that is intrinsically disordered in isolation, peptides represent powerful tools to understand PPIs. Using the interaction between small ubiquitin-like modifier (SUMO) and SUMO-interacting motifs (SIMs), here we show that N-methylation of the peptide backbone can effectively restrict accessible peptide conformations, predisposing them for protein recognition. Backbone N-methylation in appropriate locations results in faster target binding, and thus higher affinity, as shown by relaxation-based NMR experiments and computational analysis. We show that such higher affinities occur as a consequence of an increase in the energy of the unbound state, and a reduction in the entropic contribution to the binding and activation energies. Thus, backbone N-methylation may represent a useful modification within the peptidomimetic toolbox to probe β-strand mediated interactions.
N-Methylation of backbone amides has been shown to improve the affinity, interaction specificity, solubility, membrane permeability, and proteolytic stability of peptides.23,24 However, these studies have generally focused on cyclic peptides. There are far fewer reports on N-methylation of linear peptides, and even fewer for peptides solely composed of L-amino acids.25–28N-Methylation can restrict the conformations accessible to a peptide, as it disfavours backbone conformations in the bottom-left quadrant of the Ramachandran plot, which includes the α-helical region, ϕ ≈ −60° and ψ ≈ −50°.29,30 However, it also allows access to alternative conformations by lowering the difference in stability between cis/trans amide rotamers.31,32 Nonetheless, the precise manner in which N-methylation can be used to alter the backbone conformational preferences of linear peptides, in particular understanding what determines the extent to which N-methylation favours more-extended structures, is less explored. N-Methylation also changes the hydrogen-bonding capabilities of peptides. In turn, this could alter the ability of a peptide to bind to protein targets, but it can also improve the physical properties of peptides; for instance, reducing propensities to self-assemble into amyloid-like structures,33,34 and susceptibilities to certain proteases.35 Thus, in this work, we sought to explore the extent to which backbone N-methylation might serve as a tool to inform on and modulate IDR/P binding mechanism. Using the interaction between small ubiquitin-like modifier (SUMO) and SUMO-interacting motifs (SIMs) as a model to explore the effects of peptide N-methylation, we show that whilst backbone modification of some sites abrogates binding, at others it increases the peptide–protein association rate (kon) resulting in small increases in binding affinity. For the latter, this behaviour can be rationalized as follows: N-methylation restricts the accessible peptide conformations, in effect predisposing them for target recognition. This is achieved by raising the energy of the unbound state and decreasing the activation energy (entropy) required for binding.
![]() | ||
Fig. 1 Overview of the interaction between the M-IR2 region of RanBP2 (SIM2705–2717) and hSUMO-118–97 (SUMO), and the effects of backbone modification on the strength of interaction: (a) lowest energy structure of the NMR-derived structural ensemble for SIM/SUMO (PDB ID, 2LAS), highlighting key interactions (identified using Arpeggio)36 between the parent SIM peptide and SUMO; (b) differences in potency for variant SIM peptides as determined in fluorescence anisotropy competition assays; hatched bars highlight variants for which precise IC50 values could not be measured; see ESI Fig. S1† for conditions and titration data; (c) sites of N-methylation for which detailed NMR analyses were performed (Fig. 2–5). |
On this basis, we performed a systematic backbone N-methyl scan for all 13 residues of the parent SIM peptide. Peptides were prepared using Fmoc-based solid phase peptide synthesis (see the ESI† for procedures and characterization) and their relative binding affinities to SUMO were estimated using a fluorescence anisotropy competition assay to allow prioritization of sequences for more detailed study by NMR (incl. accurate affinity determination).38 Changes in inhibitory potency were observed both within and outside the consensus hot-spot region. N-Methylation at six sites (Me-Asp2705, Me-Asn2706, Me-Glu2707, Me-Ile2708, Me-Ile2712, Me-Lys2717) led to similar or slightly improved inhibitory potency to the parent SIM peptide (Fig. 1b and S1, Table S1†). By contrast, N-methylation at five sites (Me-Val2710, Me-Ile2711, Me-Ile2713, Me-Trp2714, and Me-Lys2716) led to significantly diminished inhibitory potency, and a minor reduction in potency at the two remaining sites (Me-Ile2709, and Me-Glu2715).
Three N-methylated SIM variants with IC50 values close to that of the parent SIM peptide (Me-Ile2708, Me-Ile2712, Me-Lys2717) and two variants with less potent IC50 values (Me-Val2710 and Me-Lys216) were selected (Fig. 1c) to interrogate the determinants of binding affinity, relative to the parent SIM peptide. Using one-dimensional 1H NMR experiments, we ruled out changes in the population of the cis isomer at the N-methylated amide bond and changes in peptide oligomeric state as drivers of the observed changes in potency (Fig. S2 and S3†).
Simulated annealing calculations of N-methylated peptides indicated that methylation of backbone amide bonds could restrict the accessible phi (ϕ) and psi (ψ) angles, rendering the α-helical region of Ramachandran space inaccessible (Fig. 2a). As the parent SIM sequence has been shown to lack stable secondary structure in the absence of a binding partner,38 it is possible that some of the differences in binding affinity observed could be explained by changes to the secondary structure propensity of N-methylated SIM variants. NMR chemical shifts of backbone nuclei (Hα, Cα, and Cβ) can be used as reporters of even small changes in secondary structure of disordered proteins/peptides,40 and α-like or β-like chemical shifts are indicative of an increase in the population of those secondary structures. Therefore, the backbone and side chain chemical shifts of parent SIM and its N-methylated variants were assigned using 1H–1H TOCSY, 1H–1H NOESY spectra, and natural abundance 1H–13C HSQC spectra. As anticipated, the backbone chemical shifts of the parent SIM were consistent with those for a fully unstructured (random coil) peptide (Fig. S4†). Comparison of chemical shifts within the N-methylated region of each variant, relative to parent SIM, is complicated by the fact that N-methylation will increase the electron density of the associated amide bond, due to the electron-donating character of the methyl group. In the absence of any structural changes, we expect the chemical shift of neighbouring HA atoms (i.e., HA of the N-methylated residue and HA of the preceding residue) to be shifted downfield when compared to parent SIM chemical shifts. This is what is observed for Me-Val2710, Me-Ile2712, Me-Lys2716, and Me-Lys2717 (Fig. 2b and S5†). Excluding atoms whose chemical environment is directly impacted by the introduction of the N-methyl group (i.e., atoms within six bonds of the methyl carbon), the measured backbone chemical shifts of all five N-methylated variants differed very little from the shifts of the parent SIM, indicating that no long-range secondary structural elements (e.g., extended, β-rich structure) had been detected for these unbound peptides. In further support of this conclusion, we observed no NOE's indicative of helical or strand conformations.
![]() | ||
Fig. 2 Conformational analysis of N-methylated peptides: (a) Ramachandran plots of an N-methylated residue (i) and the preceding residue (i-1) in a peptide, as a function of the repulsion energy as calculated by XPLOR-NIH (see ESI Materials and Methods†); (b) Hα or Cα secondary chemical shifts for SIM and N-methylated variant peptides. Propensity for β-strand/α-helix is shown by red/blue bars respectively with threshold for significant propensity denoted by dashed grey line. Chemical shift values around the methylation site are shown as open bars; (c) 1H–1H NOESY strips of the Me-Ile2708 variant peptide (5 °C, 500 μM peptide, 20 mM sodium phosphate, pH 7.4, 0.02% NaN3, mixing time 500 ms) highlighting the E2706-Hβ to E2709-HN NOE. |
For Me-Ile2708, the HA chemical shift pattern surrounding the N-methylated peptide bond differs for both the trans and cis isomers; the residue before the N-methylated amide is significantly more upfield than in the other SIM variants, while the residue after the N-methylated amide is more downfield (Fig. 2b), suggesting the possibility of a turn-like collapse.41,42 An NOE between N2706-HA and E2709-HN was also observed for the trans isomer of Me-Ile2708 (Fig. 2c), indicating some localized structuring of residues at the N terminus of this variant. However, we note that the N terminus does not participate in SIM/SUMO recognition and thus it is difficult to predict how such ordering will affect binding.
Overall, our NMR analysis on the unbound peptides strongly indicates that restriction of the available conformational space as a consequence of N-methylation does not induce significant changes in secondary structure. Thus, the observed changes in affinity likely arise from altered binding kinetics or differences in the bound SIM/SUMO structure.
![]() | ||
Fig. 4 (a) 15N CPMG profiles of 15N-SUMO in the presence of Me-Lys2716 at 750 MHz (purple) and 600 MHz (green). Experimental data are shown as dots, while fits to a 2-state model (see Fig. 3a) are shown as solid lines. (b) Correlation of fitted 15N Δω values for the Me-Lys2716 bound SUMO with those experimentally observed for SUMO bound to parent SIM (Pearson correlation coefficient ∼ 0.86). |
Peptide | K D at 5 °C (μM) | k on (s−1 M−1) × 105 | k off (s−1) |
---|---|---|---|
Me-Ile2708 | 14.0 ± 0.4 | 3.27 ± 0.09 | 4.58 ± 0.04 |
Me-Val2710 | 388 ± 7 | 5.36 ± 0.10 | 208 ± 1 |
Me-Ile2712 | 14.5 ± 0.9 | 4.10 ± 0.28 | 5.90 ± 0.08 |
Me-Lys2716 | 510 ± 40 | 10.9 ± 0.7 | 553 ± 17 |
Me-Lys-2717 | 13.6 ± 0.5 | 6.57 ± 0.27 | 8.95 ± 0.12 |
Parent SIM | 33.6 ± 0.9 | 2.45 ± 0.07 | 8.22 ± 0.08 |
For the weak peptide binders (Me-Val2710 and Me-Lys2716), only the unbound SUMO resonances were observed in 1H–15N HSQC spectra, even at high SIM concentrations. This suggests that the population of the bound complex is small and/or that binding does not take place in the slow chemical shift timescale. Thus, to investigate the binding kinetics of those variants, we used Carr–Purcell–Meiboom–Gill (CPMG) relaxation dispersion that is sensitive to exchange between states with skewed populations on the millisecond timescale.43 Large CPMG profiles for SUMO residues in the SIM-binding site were observed at 5 °C upon addition of Me-Val2710 or Me-Lys2716 (Fig. 4a and S12–S14†). These profiles were absent for apo SUMO (Fig. S15†), suggesting that the observed millisecond dynamics are due to transient SIM binding. Global fitting of CPMG data at 600 and 750 MHz to a 2-state model yielded excellent fits (Fig. 4, S12 and S13†). CPMG-derived chemical shifts of the transiently populated, bound state of SUMO in the presence of Me-Val2710 or Me-Lys2716 correlated well with those of the stable, parent SIM-bound state, with the exception of some residues in the SIM binding pocket (see Fig. S12b and S13b†), suggesting that the bound state of SUMO in the presence of these SIM variants is not significantly different to that with parent SIM. However, koff values or the weakly bound variants increased by more than 50-fold (∼200–550 s−1; Fig. 5b) – therefore, while the structure of SUMO in the bound state complex is unchanged by N-methylation, N-methylation at residues Val2710 and Lys2716 either prevents or hinders the ability of the SIM peptide to adopt its canonical binding conformation, leading to fast dissociation and a lower affinity interaction. It is surprising that Me-Val2710 does not show improved binding affinity as the amide NH is not involved in hydrogen-bonding and methylation should restrict Ramachandran space to conformations that favour binding.44 Therefore, it is likely that methylation adversely influences the accessible χ space of the isopropyl side chain of Val2710 (a hot-spot residue), making it incompatible with SUMO binding. In the case of Lys2716, we consider it more likely that methylation of this residue perturbs the allowable χ space for Trp2714.
All together, our results suggest that the restriction of the conformational space available to N-methylated peptides can increase association kinetics which correlates with enhanced binding. Conversely, in some cases, if the modified peptide cannot adopt a stable binding conformation, N-methylation can favour dissociation which correlates with diminished binding.
ΔH (kJ mol−1) | TΔS (kJ mol−1) | ΔG (kJ mol−1) | |
---|---|---|---|
Parent SIM | |||
Binding (KD) | 113.1 ± 0.7 | 146.7 ± 2.8 | −33.6 ± 1.0 |
Association (activation, kon) | 164.8 ± 1.0 | 134.6 ± 1.0 | 30.2 ± 1.2 |
Dissociation (activation, koff) | 52.0 ± 0.3 | −11.9 ± 0.3 | 64.0 ± 0.4 |
![]() |
|||
Me-Ile2712 | |||
Binding (KD) | 125.2 ± 0.2 | 161.8 ± 2.0 | −36.6 ± 2.0 |
Association (activation, kon) | 172.5 ± 2.1 | 145.5 ± 2.1 | 27.5 ± 2.2 |
Dissociation (activation, koff) | 50.2 ± 0.3 | −14.7 ± 0.2 | 64.8 ± 0.4 |
![]() |
|||
Me-Lys2716 | |||
Binding (KD) | 31.4 ± 0.2 | 52.5 ± 0.2 | −21.1 ± 0.3 |
Association (activation, kon) | 48.5 ± 0.2 | 13.0 ± 0.2 | 35.4 ± 0.3 |
Dissociation (activation, koff) | 17.9 ± 0.1 | −38.5 ± 0.1 | 56.5 ± 0.1 |
Eyring plots were obtained for the temperature-dependence of the kon and koff rates (Fig. 5d and e), from which the association/dissociation activation enthalpies (ΔH‡ass/ΔH‡diss), activation/dissociation entropies (TΔS‡ass/TΔS‡diss), and thus activation/dissociation free energies (ΔG‡ass/ΔG‡diss), could be determined. The association of SIM and its variants with SUMO has an enthalpic activation barrier in all cases (Table 2), while peptide dissociation has an enthalpic barrier for parent SIM and Me-Ile2712, and a smaller entropic barrier for Me-Lys2716 (Fig. 5d, Table 2). At 25 °C, TΔS‡ass for Me-Ile2712 is more favourable by ≈11 kJ mol−1, which is compensated only by a ≈8 kJ mol−1 more unfavourable ΔH‡ass. Considering that the dissociation free energy barrier, ΔG‡diss, is practically identical for the parent and Me-Ile2712 peptides (≈64 kJ mol−1), the slightly more favourable TΔS‡ass gives rise to the 3 kJ mol−1 decrease in ΔGbind for Me-Ile2712 (Fig. 5c–e). Taken together, these data suggest that entropy-driven lowering of the association activation barrier arising from an increase in the free energy of the N-methylated peptide (relative to the parent SIM) represents a plausible explanation for the increased affinity for Me-Ile2712.
In setting these results within the context of potential molecular mechanisms of recognition, we note the following additional considerations. First, there are likely subtle effects, such as a small increase in hydrogen-bond accepting ability of the carbonyl that might be expected upon N-methylation of the peptide bond. In turn, this would be anticipated to increase binding enthalpy. In addition, we cannot exclude the possibility that increased hydrophobicity, differential solvation of the methylated peptides, or subtle changes in bound conformation influence binding kinetics and affinity. Nonetheless, the entropy-driven increase in kon rates that we observe lead us to conclude that increased affinities are caused by restricting the accessible conformational space of the N-methylated peptides. We simulated Ramachandran plots, which show that N-methylation significantly limits the phi(ϕ)/psi(ψ) angles accessible to residues on either side of the methylation site, such that residues are limited to extended or turn-like conformations (i.e., excluded from α-helical space). However, NMR analyses suggest the unbound peptides do not adopt a defined conformation in the absence of SUMO. We contend that the overall ensemble of SIM conformers has a higher ground state energy and that this lowers the entropy of activation for SUMO binding (see free energy diagrams in Fig. 6a). We note that the polyproline-II conformation is somewhat intermediate between the α-helix and β-strand conformations,46 and that N-methylation is known to restrict the conformational space of peptide backbones.29,30 Thus, whilst the methylated peptides cannot be considered as pre-organized for SUMO binding, the ensemble is expected to disfavour α-space and thus favour β-space localized around the N-methylated residue so is primed or predisposed towards SUMO recognition. Previously, pre-organization of a peptide that recognises its target through a bind-and-fold20 mechanism (Fig. 6b) was shown not to enhance affinity for its target, because constraining limits “the number of ways to bind”,47 whilst for a peptide which recognises its target through conformational selection (Fig. 6b),20 constraining should increase affinity.13 Given N-methylation does not seem to induce a specific extended conformer and instead favours an ensemble of conformers that are compatible with binding, the effect observed here may represent a useful strategy to modulate thermodynamic and kinetic parameters of binding for ligands which bind their target through conformational selection or bind-and-fold mechanisms. Modulating peptide binding by tuning the entropy of activation/binding represents an untapped approach for design of peptidomimetic ligands. Whilst the overall effects on affinity observed here are small, methylation is known to confer proteolytic resistance and improved permeability, thus employing this modification within the toolkit for optimizing peptidomimetics may prove useful for development of therapeutically relevant PPI inhibitors. Establishing design rules that predict where an N-methyl group should be placed will require a larger data set; excluding amide NH's involved in hydrogen bonding or where a steric clash with the protein target would be envisioned to reduce the number of N-methyl variants that should be explored, however our results imply that N-Methylation may sterically influence local (Val2710) and remote (via Lys2716) side chain orientation to adversely impact target binding. Our future studies will explore backbone N-methylation to mechanistic analyses of other β-strand mediated PPIs and to explore the generality of our observations.
![]() | ||
Fig. 6 (a) Potential free energy diagrams together with schematics that could explain the data presented in Table 2. Grey, green and magenta dashed lines represent ΔGbind, ΔG‡ass, and ΔG‡diss, respectively, and their values are given in kJ mol−1 at 25 °C (note: we show the parent and Me-Ile2712 bound complexes as isoenergetic on the basis of their HSQC bound state spectra and ΔG‡diss); (b) schematic illustrates the extremes of conformational selection and bind-and-fold protein binding mechanisms. |
Footnotes |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4sc02240h |
‡ Current address: The Francis Crick Institute, 1 Midland Rd, London NW1 1AT, UK. |
This journal is © The Royal Society of Chemistry 2024 |