The cystine knot promotes folding and not thermodynamic stability in vascular endothelial growth factor*

Cystine knots consist of three intertwined disulfide bridges and are considered major determinants of protein stability in proteins in which they occur. We questioned this function and observed that removal of individual disulfide bridges in human vascular endothelial growth factor (VEGF) does not reduce its thermodynamic stability but reduces its unexpected high thermal stability of 108 degrees C by up to 40 degrees C. In wild-type VEGF (deltaG(u,25)(0) = 5.1 kcal.mol(-1)), the knot is responsible for a large entropic stabilization of TdeltaS(u,25)(0) = -39.3 kcal mol(-1), which is compensated for by a deltaH(u,25)(0) of -34.2 kcal mol(-1). In the disulfide-deficient mutants, this entropic stabilization disappears, but instead of a decrease, we observe an increase in the thermodynamic stability by about 2 kcal.mol(-1). A detailed crystallographic analysis of the mutant structures suggests a role of the cystine knot motif in protein folding rather than in the stabilization of the folded state. When assuming that the sequential order of the disulfide bridge formation is conserved between VEGF and glycoprotein alpha-subunit, the crystal structure of the mutant C61A-C104A, which deviates by a root mean square deviation of more than 2.2 A from wild-type VEGF, identifies a true folding intermediate of VEGF.


SUMMARY
Cystine knots consist of three intertwined disulfide bridges and are considered major determinants of protein stability in proteins where they occur. We questioned this function and observed that removal of individual disulfide bridges in human vascular endothelial growth factor (VEGF) does not reduce its thermodynamic stability but reduces its unexpected high thermal stability of 108 °C by up to 40 °C. In wild-type VEGF ( ∆ =5.1 kcal · mol -1 ) the knot is responsible for a large entropic stabilisation of = -39.3 kcal mol -1 , which is compensated for by a of -34.2 kcal mol -1 . In the disulfide deficient mutants this entropic stabilisation disappears, but instead of a decrease, we observe an increase in the thermodynamic stability by about 2 kcal · mol -1 . A detailed crystallographic analysis of the mutant structures suggests a role of the cystine knot motif in protein folding rather than in the stabilisation of the folded state. When assuming that the sequential order of the disulfide bridge formation is conserved between VEGF and glycoprotein α-subunit, the crystal structure of the mutant C61A-C104A, which deviates by an r.m.s. deviation of more than 2.2 Å from wild-type VEGF identifies a true folding intermediate of VEGF.

INTRODUCTION
The cystine knot structural motif consists of three highly intertwined disulfide bridges and occurs in the small inhibitor cystine knot proteins, the cyclotides and the growth factor cystine knot proteins. The first two classes of proteins consist of small polypeptides ranging from 25 to 30 amino acids with functions as diverse as protease inhibition (squash family of inhibitors), ion channel blocking (toxins) and antimicrobial activity (cyclotides) (1)(2)(3). The third class consists of the growth factor cystine knot proteins, with members such as TFG-β, α−NGF, PDGF, VEGF, BMPs, IL-17s and many others. These physiologically important growth factors share a common monomer structure but differ in their mode of dimerisation (3), thus introducing molecular diversity into the structural mechanism by which these growth factors activate their cognitive receptors (4). Whole genome analyses hint that the cystine knot motif might be very common in extracellular signalling molecules (5).
The molecular function of the cystine knot motif is poorly understood. In this motif, two disulfide bridges connect two neighbouring chain segments and form a ring structure; a third disulfide bridge penetrates this ring segment and crosslinks two additional chain segments. By implication the cystine knot motif interlocks four separate chain segments.
Considerations on the small size of the inhibitor cystine knot proteins and the lack of extended hydrophobic core regions in both the inhibitor and growth factor cystine knot proteins, lead to the assumption that this structural motif is a major determinant for the thermodynamic stability of these proteins. Thus, amide-exchange experiments estimated the thermodynamic stability of the toxin ω-MVIIA to be about 4 kcal mol -1 (6). This equals the stabilisation often introduced by single disulfide bridges of 3 to 5 kcal mol -1 (7,8) and therefore it is plausible that removal of single cystines impairs the stability of these proteins.
We challenged this assumption and studied the thermodynamic and structural properties of wild-type vascular endothelial growth factor (VEGF) and of three of its four -4 -by guest on April 17, 2019 http://www.jbc.org/ Downloaded from possible cystine deletion mutants. We produced the mutants C51A-C60A' with the intermolecular disulfide bridge removed (named hereafter ∆I), and C57A-C102A (∆II) and C61A-C104A (∆III) in which the disulfide bridges forming the outer ring of the knot motif have been removed (Fig. 1A). Unexpectedly we observe that none of these mutants appears to be thermodynamically destabilised. Moreover, deletion of the cystine knot disulfide bridges increases the thermodynamic stability by ~ 2kcal · mol -1 . In contrast we observe that individual disulfide bridges are to various extents crucial for the structural integrity and thermal stability of VEGF.
Protein concentrations were determined spectrophotometrically at 276 nm using absorption coefficients of A 1%,1cm = 5.74 for wild-type VEGF and A 1%,1cm = 5.64 for the VEGF variants ∆I, ∆II and ∆III calculated from the amino acid composition (10). The CD spectra in the far ultraviolet region (190 to 260 nm) were measured with a Jasco-J720 spectropolarimeter and are characteristic for β proteins with a minimum at 212 nm. The far ultraviolet CD spectra   (15) as a search model and refined to convergence with program CNS (16). In case of mutant ∆III no unambiguous molecular replacement solution could be obtained. However, ∆III crystallises in the same spacegroup and with similar cell dimensions than ∆I and therefore we assumed that the molecular packing in the ∆I and ∆III crystals is similar. Phases for the ∆III dataset were calculated to 4.0 Å with the ∆I model and extended in 1000 steps to 2.5 Å by two-fold noncrystallographic symmetry averaging and solvent flattening using program DM (14). The redefinition of the NCS-symmetry operator and the repeating of the density modification procedure yielded electron density maps, which could be readily extended to 1.32 Å using the free atom refinement protocol of program WARP (17). Conventional refinement together with manual inspection was performed with programs REFMAC (14) and ONO (18). In the final refinement round anisotropic thermal displacement factors were introduced. Consequently, the free R-factor dropped from 22.6 to 19.5 % and in parallel the crystallographic R-factor from 20.6 to 16.5 %.

RESULTS AND DISCUSSION
Thermodynamic and thermal stability of wild-type VEGF and mutants -The normalized transition curves of the guanidine-hydrochloride (GdnHCl) induced unfolding of wild-type VEGF and variants ∆II and ∆III are shown in Fig. 2A (19). The good agreement between the free energies obtained by DSC and GdnHCl-induced unfolding supports our approach (Table 1). This agreement and values close to one for κ = ∆Η vH /∆Η cal (quotient between the van't Hoff enthalpy ∆H vH and the calorimetric enthalpy ∆H cal ) also corroborates the assumption of a two-state unfolding mechanism (20).
The discrepancy observed between the high thermal stability ( T = 108.7 °C) and the moderate free unfolding energy ( = 5.1 kcal mol -1 ) of wild-type VEGF is unusual because in most proteins high thermal stability correlates with large thermodynamic stability (21)(22)(23)(24). Thus the -value of wild-type VEGF is reminiscent of proteins from thermophilic organisms and unexpectedly high for a human protein. Removal of a single disulfide bridge from the cystine knot lowers the T m value by as much as 38.9 °C (∆II,  Therefore a smaller negative T is observed for ∆III and a value more common in proteins (26) is observed for ∆II ( Table 1). The observed increase in ∆C P and the unfolding cooperativity m of the mutants reflect this increase in the conformational freedom and the concomitant increase in the protein accessible surface upon unfolding (27,28).

S ∆
We infer that none of the disulfide bridges increases the thermodynamic stability of VEGF -Our analysis of the thermodynamic role of the disulfide bridges in VEGF must appear incomplete. In case of mutant ∆I, the breakdown of the two-state approximation paired with protein aggregation prevented its characterisation. Nevertheless from the fact that ∆I can be refolded and crystallised at room temperature (see below) we estimate that its free unfolding energy can not be considerably smaller than the already small thermodynamic stability of wild-type VEGF. Indirect evidence hints that this might also be the case for the fourth possible mutant, which we failed to produce. The recent crystal structure of IL-17E (29) revealed that IL-17s belong to the cystine knot growth factor family with the particularity that the disulfide bridge penetrating the ring structure of the cystine knot is missing. The conclusion that this disulfide bridge is not a prerequisite for the thermodynamic -10 -  As an indicator for changes in the dimerisation of VEGF we studied the coplanar arrangement of the monomers in the dimer. For each variant we calculated the planarity angle.
It is defined as the deviation from 90° of the angle enclosed between the line connecting the Cα's of Glu30 and Ser74 in each monomer and the two-fold symmetry axis relating the monomers in the dimer ( parallel to the two-fold axis, when comparing the wild-type structure to ∆III (Fig. 1C). This appears a direct consequence of rearrangements in the side-chain orientations in the dimer interface in this mutant (see below).
In addition to global changes, we observe a number of local structural changes in the mutants ( Table 3, Fig. 3). In ∆II only small changes occur. When replacing Cys57 and Cys102 with alanines, the distance between the equivalent Cα's is slightly reduced from 3.97 to 3.59 Å; any additional changes lay well within the deviations observed for different wildtype VEGF structures (15). In ∆I the distance between the equivalent Cα's is increased from 3.87 to 6.84 Å ( Table 3, Fig. 3A). This reflects a widening of the monomer-monomer interface at the position where the intermolecular disulfide bridge occurs in wild-type VEGF.
As a consequence the cystine-knot motif rotates as a rigid body away from the interface. The structure of the cystine knot itself is not perturbed.
More drastic changes occur when replacing the disulfide bridge formed between Cys61 and Cys104 with alanine residues (∆III). Superposition of the ∆III monomer onto wildtype VEGF yields an r.m.s. deviation as large as 1.99 Å. This is mainly caused by a conformational change in the loop segment connecting Cys60 and Cys68. Ala61 points away from Ala104 and the distance between the Cα-positions is increased from 3.81 to 12.14 Å (Fig. 3C, D). In ∆III the diameter of the VEGF dimer is reduced by about 2 Å whereas this distance does barely change in ∆I and ∆II (Table 3). This is only partly a consequence of a -12 -  (Table 3). More importantly, we observe a tighter packing in the dimer interface, reducing by 1.7 Å the distance between the two β-strands lining the centre of the interface (distance calculated between the equivalent Cα-positions of Glu30). Two novel symmetry equivalent salt bridges are formed across the interface between His27 and Glu30 and reorientations in the side-chains of residues Ile29 and Leu32 occur. This increase in compactness and in the packing of the interface is totally unexpected. The disulfide bridge, removal of which, causes these changes, is of all the disulfide bridges studied the most remote from the centre of the dimer.
It is not possible to directly correlate the observed structures with the thermodynamic unfolding data because with the exception of an increase in ∆C P and m, which reflect an increase in the solvent accessibility of the unfolded state, we have not characterised the unfolded state structures. We only note that the mutant ∆II, which structure is the most similar to wild-type VEGF shows the most drastic changes in T m , ∆C P , m, . Therefore the altered thermodynamic properties of this mutant must result largely from the effect of these mutations on the unfolded state. Previous studies have revealed the importance of the disulfide bridges for the folding of glycoprotein hormone α-subunit in vivo (30). Here we showed that in vitro it is possible to generate cystine deficient mutants of the cystine knot growth factor VEGF and to subsequently investigate their crystal structures. This was possible because, as we showed, individual disulfide bridges in VEGF only contribute to the thermal stability and structural integrity of the protein but they do not change its thermodynamic stability. This property might be of primary importance with respect to protein folding. The close spatial proximity of -13 - The crystal structures show that the cystine knot is of primary importance for the structural integrity of VEGF. Although the knot appears to only affect the fine-structure of the protein, we anticipate that these changes have drastic consequences for the biological activity of VEGF. Loop regions adjacent to the knot participate in receptor binding and in particular residue Glu64 located on the surface loop formed between residues 61 and 68 determines VEGF receptor specificity (37,38). The conformation of this loop is drastically changed in mutant ∆III (Fig. 3C, D) and the spatial separation between residues participating in receptor binding is also changed in ∆I (15,34). Therefore the biological activity profile should be altered in these mutants.
-14 -    -20 -  The deviation from planarity gives a measure for the coplanarity of the monomers.