Assigning the NMR spectrum to the specific protons of a protein is the prerequisite to map interactions and any conformational change. At 30°C, the 2D ¹H -¹⁵N HSQC of mNGF is optimal and reveals a wide dispersion of signals characteristic of proteins with a predominantly β-sheet content which is consistent with the X-ray structure. The indole correlations of the three Trp residues are clearly observable in the distinctive and typically isolated region of the spectrum around 10.5 ppm (¹H) and 135 ppm (¹⁵N). All Gln and Asn side chains are detected. We achieved virtually full assignment of the spectrum. Conversion of the NOE information into a structural model was not trivial because the potential symmetry (two-fold) of the homodimer makes it hard to distinguish between intra- and inter-molecular NOEs. The problem, which has been debated for years (Saudek et al., 1991; Nilges and O'Donoghue, 1998), was circumvented thanks to the software support of ARIA (Rieping et al., 2007) which allows discrimination of intra- from inter-molecular NOEs and careful iterative analysis of the violations (Table 1). The process led eventually to a well-converged ensemble (Figure 1A) with a root mean square deviation (r.m.s.d.) of 1.3 Å from the structure with the lowest global energy as calculated on 236 residues (Figure 1B). It closely resembles the available X-ray structures, especially in the Cys-knot, while exhibiting larger variability in the loop regions (Figure 1C). The r.m.s.d. values between the structure with the lowest global energy and the crystal structures of unbound mNGF (PDB ID 1BTG, protomers B,C) or NGF bound to lipids (PDB ID 4EAX or 4XPJ) for the backbone atoms of the core residues (15–22, 51–58, 78–89, 100–111) which exclude the loops are 2.32, 2.14, and 2.12 Å respectively. The structure in solution allows us to address a number of crucial aspects as detailed in the following sections.

From our data, we can obtain direct information on the structure of the N-terminus (residues S1-M9) from the NOE patterns. In solution, the N-terminus, whose resonances can be observed distinctly, does not adopt a regular secondary structure. However, a number of medium-range NOEs were observed between the backbone and side-chains of these residues, suggesting the existence of a short nascent helical turn (residues 6–10). The average O_i…NH_i+3 and O_i…NH_i+4 distances between residues S2-G10 were measured for the 20 lowest global energy models to check the presence of distances compatible with hydrogen bonds. Overall, the V6-G10 segment has a clear propensity to form a short helix, the O_i…NH_i+3 and O_i…NH_i+4 distances being within 2.6 and 6.1 Å over the whole bundle. In almost 50% of the 20 lowest global energy models, the V6-G10 segment adopts either a 3₁₀ helix or a α-helix conformation (Figure 1A, inset). These distances are within 2.1 and 6.2 Å in the lower global energy model exhibiting a helical conformation. These results provide a clear answer to whether and how much the N-terminus is folded and clarify its tendency to fold into a helical conformation. The secondary chemical shifts, that is the difference between the observed values and the random coil values for the same amino acid, confirm a helical tendency of this region (data not shown).

While assigning the NGF spectrum which was troublesome due to significant overlap, it became clear that some interesting dynamical processes were going on. At 30°C the vast majority of the peaks have homogeneous intensity but a limited set of resonances of weaker intensity are observed (Figure 2A). After virtually full spectral assignment, these weaker resonances were found to be duplicate assignments for a small set of specific resonances which cluster around loops I and II and on the stem region (region IV-loop V) that points toward loop II (residues 39–47) (Figures 2B,C). Notably, W99, present in this region, shows doubling of the indole proton. Loop II, the region where most of the resonance splitting is observed, is also involved in small ligand binding (PDB ID 4EAX and 4XPJ). Peak doubling was furthermore observed for the side chains of a few residues in the ¹³C HCCH-TOCSY and ¹³C NOESY-HSQC spectra (data not shown). To test for the presence of conformational exchange, ¹⁵N HSQC spectra were recorded at 10°C intervals between 10°C and 40°C (Figure 2A). At 10°C, the peak intensity is highly heterogeneous with only ca. 30% of the number expected for a homodimer of 118 amino acids. By progressively increasing the temperature to 40°C, all the expected signals return, with the vast majority showing homogeneous intensity. However, a subset of secondary species of lower intensity remained observable. The overall quality of the spectrum is consistent with that expected for a homodimer of 26 kDa, with a line width indicative of the molecular size of the system. For a homodimeric protein, resonance doubling can be explained in at least two possible ways. It could indicate an intrinsic local asymmetry of the homodimer over the NMR timescale. This possibility can, however, be ruled out since the most affected region around loop II is not engaged in the homodimer interface. Alternatively, it could arise from the presence of multiple conformational species in equilibrium on the slow-exchange timescale. This second possibility seems more likely given the dramatic change in the number and in the relative peak intensities observed in the ¹⁵N HSQC as a function of temperature.

The data thus suggest that the NGF loops adopt different conformations in the slow exchange regime in the NMR timeframe.

We measured the T1, T2 and heteronuclear NOE (hetNOE) parameters to obtain information on the dynamics in solution. The experimental correlation time (T_C) is 14.8 ns which is consistent with the expected molecular weight of the dimer. The T1, T2, and NOE values correlate well with the secondary structure (Figure 3A) and highlight the more rigid β-sheet regions. The loops, especially II and V, and the C-terminus are relatively more flexible showing hetNOE values lower than the average (0.7). However, overall, the hetNOE profile is relatively flat with some minor exceptions which could well be explained by residue overlap.

Intrigued by these results, we ran a molecular dynamics simulation of mNGF (200 ns) to further explore the tendency of the N-terminus to adopt a helical conformation and the conformational space of the loop regions. The ensemble of structures sampled in the MD trajectory is variable, but in line with the NMR analysis (Figure 3B). Not having imposed any symmetry constraints we see of course a different behavior in the two protomers. Over the simulation time scale, all residues in the tract His4-Phe7 spend a significant percentage of time in a 3₁₀ or α-helix conformation at least in one of the protomers (Figure 3C). Residues H4-G10, in both protomers, unwind and fold during the simulation but are able to form transient hydrogen bonds mostly between the carbonyl oxygen of H4 and the amide hydrogen of F7 and the carbonyl oxygen of P5 and the amide hydrogen of H8. The carbonyl oxygen of V6 is also engaged only in one protomer in hydrogen bonds with the amide hydrogen of either M9 or G10 residues. The main-chain–side-chain interactions between residues H4/F7, V6/M9, and F7/G10 are overall more preserved than the side-chain–side-chain interactions between residues P5/H8 and V6/M9 as well as the interaction between the residues V6/G10. The MD results are in line with the observed NOE contacts supporting that in at least one of the protomers (Table S1), a short segment within the central region of the mNGF N-terminus adopts a α-helix (15%) or a 3₁₀-helix (35%) conformation for a small but significant fraction of time (Figure 3D). According to these conclusions, a previous MD study on an isolated linear peptide spanning the N-terminal sequence (Stanzione et al., 2010) has shown that most of the adopted conformations do not present a regular secondary structure but a helical structure can form for a limited fraction of conformers in the ensemble.

We can thus conclude that the NGF N-terminus comprises a nascent helix which can be further stabilized upon binding with its partners.

The C_α r.m.s.d. along the MD trajectory shows that, besides the N- and C-termini, the regions encompassing loops II and V are comparably more flexible and exhibit large-scale conformational flexibility (Figure 3B). The loop variations are however relatively small as compared to the flexibility of the N- and C-termini. This indicates that the loops are plastic but not flexible in agreement with the differences observed in these regions in the crystallographic structures of apo mNGF (PDB ID 1BET, 1BTG) and in complex with lysophosphatidylserine (PDB ID 4EAX) and lysophosphatidylinositol (PDB ID 4XPJ). This information is of crucial importance for the development of small organic compounds (Brahimi et al., 2014) and peptidomimetics based on the combination of different NGF loop fragments (Colangelo et al., 2008). We thus analyzed the relative intra- and inter-protomer distances between loops I, II, III, and V (Figure S1). The shortest and the longest inter-protomer C_α distances between the Asn46 of Loop II are 11 Å (apo form PDB ID 1BTG, protomers B,C) and 30 Å respectively (holo forms PDB ID 4XPJ, protomers A,B and PDB ID 4EAX, protomers A,B and C,D) indicating large scale motions. The corresponding distances, in the most representative MD structure (the most populated cluster center) and in the lowest global energy NMR structure are 13 and 16 Å respectively. The longest and shortest inter-protomers distances between Lys95 of Loop V are 42 Å (apo form PDB ID 1BTG, protomers B,C) and 40 Å (holo form PDB ID 4XPJ). The corresponding distances in the most representative MD structure (the most populated cluster center) and in the lowest global energy NMR structure are 34 and 37 Å respectively (Figure S1).

For comparison, in the crystal structures of free and bound NGF, the conformation of loop II has high plasticity with an overall opening of the structure upon the accommodation of the small ligands (Figure 1C). It is tempting to speculate that the conformational equilibrium experienced by loop II and V, supported also by resonance doubling, is functionally relevant for the biological activity of the molecule, including interactions with the binding partners, i.e. the receptors TrkA and p75^NTR. These features highlight the differences between NGF and other homodimeric β-sandwich structures.

We can thus conclude that NGF has long plastic but relatively rigid loops. This information is of crucial importance for drug design (Bannwarth and Kostine, 2014).

Finally, we used the information gained in solution to study the interaction of NGF with αD11, one of the well-characterized antibodies. We titrated ¹⁵N-labeled NGF with αD11 using the MAb protein. We used sub-stoichiometric amounts of αD11 to allow the observation of shift changes, given that the high affinity expected and the large molecular weight of the complex could prevent observation of the signals at stoichiometric concentrations (Figure 4A). The most appreciable perturbations in the NGF spectrum occur in three distinct sets of resonances which disappear (V14, G23, F101, C110, V111), decrease in intensity (T26, L39, N45, N46, Y52, E65, D93, W99, R100, K115) or shift significantly (G10, S13, V18, A40, F49, T56, Y79, T81, T92, E94, K95, Q96, I102, V109) indicating slow, slow-intermediate and fast-exchange regimes, respectively. When mapping these regions onto the NGF structure, a set of the affected residues (V18, G23, T26, L39, A40, F49, Y52, D93, Q96, W99, R100, F101) overlap with those previously identified in the SAXS hNGF-αD11 FAb complex model (Covaceuszach et al., 2008). Additionally, residues G10-V14, V64-I71, T81, and V109-V111, which cluster close together in space, suggest a previously unidentified epitope (Figure 4B). Interestingly, this region is not included in the FAb-NGF contacts in the crystal structures of the complexes of hNGF with the anti-NGF antibodies Tanezumab (PDB ID 4EDW) and its murine precursor MAb 911 (PDB ID 4EDX), whose epitopes are otherwise comparable to those of αD11 FAb. Thus, the solution NMR epitope mapping of the MAb αD11 unveiled a previously undetected epitope, further supporting why αD11 antibody binding differs significantly between mature NGF and proNGF: in the latter the pro-peptide could hinder the accessibility of the NGF surface to the antibody.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.