Crystallization of insulin is of high importance in pharmaceutical formulations and in insulin manufacturing and has been systematically investigated since the 1920s [6
]. By introducing chaotropic agents in the crystallization experiments, we succeeded in identifying two new crystal forms of native human insulin. The crystals were found using two different crystallization experiments. The first crystals were obtained in a crystallization screen with varying concentrations of urea and sodium chloride in presence of zinc and resorcinol. The crystals were initially characterized by X-ray powder diffraction and were shown to have a powder pattern differing from previously known insulin crystal forms [21
]. Further optimizations of crystallization conditions resulted in crystals suitable for single crystal analysis and determination of crystal system which was found to be orthorhombic in space group C2221
. The crystals appeared in the pH range 6.0 – 6.5 while a second type of crystals, characterized as monoclinic with space group C2 grew in the pH range 6.5 – 7.0. In the overlapping pH interval around pH 6.5, the C2221
crystals were present in wells with lower salt and urea concentrations. A few drops contained a mixture of the two crystal types. In a parallel experiment, the urea and sodium chloride were substituted for thiocyanate. Interestingly, the same two crystal types appeared here, at the same pH intervals, with a clearer pH distinction at pH 7.0. Crystallization with thiocyanate or chloride ions without a phenol derivative has previously been shown to stabilize the T3
form of hexameric insulin in a rhombohedral crystal system [16
]. In our case when resorcinol and thiocyanate were present, the orthorhombic C2221
and monoclinic C2 crystals appeared. The well characterized monoclinic crystals in space group P21
] were present at pH values above 7.0 in wells with low salt and high urea concentration and increased in frequency as the pH was raised to become the dominating crystal form at pH ≥ 7.5. The crystals obtained in presence of urea will be referred to as C2221
urea and C2urea while the two crystal forms obtained with thiocyanate are referred to as C2221
scn and C2scn.
Structure of insulin in the orthorhombic lattice
The crystals grown at pH 6.5, from both the urea- and thiocyanate (NaSCN) screen, were found to belong to space group C2221 with unit cell parameters a = 59 Å, b = 220 Å, c = 223 Å. The asymmetric unit contains three insulin hexamers with a crystal solvent content of 64%. The hexamers have R6 conformation and contain two zinc atoms/hexamer coordinated to three histidine residues (HisB10). In the C2221urea crystal, the zinc is additionally coordinated by a chloride ion at an average distance of 2.15 ± 0.10 Å, whilst in the C2221scn crystal structure the chloride ion is replaced by a thiocyanate. The thiocyanate coordinates to zinc through its nitrogen atom with an average distance of 1.82 ± 0.04 Å. The three hexamers in the asymmetric unit are arranged in an angular formation where the central hexamer connects the two outer hexamers with an angle of ~110°, Figure . The non-crystallographic three-fold axes which pass through the two zinc atoms in each hexamer are almost orthogonal to each other. In both the C2221urea and C2221scn structures, one co-crystallized resorcinol molecule is bound to each insulin monomer in the phenolic binding pocket. The resorcinol molecule is hydrogen bonded with its first hydroxy group to the carbonyl O atom of CysA6 (average distance 2.6 Å), and the N atom of CysA11 (average distance 2.9 Å). The second hydroxy group hydrogen bonds to a water molecule with an average distance of 2.7 Å. This water molecule forms another hydrogen bond to the O atom of CysA11 with an average distance of 2.8 Å. In the final stages of refinement, one glycerol molecule was modeled into the C2221scn structure at a position where it interacted through its oxygen atoms with the amide nitrogen of PheB1 (2.9 Å) and the carbonyl oxygen of ThrA8 (2.9 Å). The crystal packing of the C2221urea structure is shown in Figure .
Figure 1 Crystal packing. (a) The asymmetric unit of the C2221 crystal form viewed down the non-crystallographic three-fold axis of the central hexamer. The flanking hexamers are located around the central hexamer at an angle of ~110°. The local non-crystallographic (more ...)
Structure of insulin in the monoclinic lattice
The crystals obtained at slightly higher pH (pH 7.0) belong to the monoclinic space group C2 with cell dimensions a = 100 Å, b = 60 Å, c = 62 Å, β = 116°. They contain one hexamer with R6
conformation in the asymmetric unit and have a solvent content of 50%. The crystal packing is shown in Figure . Both the C2urea and C2scn structures have two zinc atoms/hexamer located 14.7 Å and 15.2 Å apart respectively. The zinc coordination is identical to the C2221
crystals. In the C2urea structure, two additional resorcinol molecules could be fitted into the electron density. The location of the first is very close to binding site II, described in [13
]. At this site, the first hydroxy group of the resorcinol molecule is hydrogen bonded to the OG atom of SerB9 (2.9 Å) in an alternating conformation. The other hydroxyl group interacts with the carbonyl oxygen of GluB13 (2.5 Å) and a water molecule (3.1 Å). The water molecule, in turn, makes a hydrogen bond to the carbonyl oxygen of SerB9 (3.1 Å). In contrast to the phenolic binding interactions observed earlier in the PDB entry 1ZEG
], where the phenolic oxygen hydrogen binds to HisB5, the angular orientation of the HisB5 in the C2urea structure does not seem to permit any interaction with the resorcinol molecule. The second additional resorcinol is located at the surface of the insulin in a solvent channel between two monomers, surrounded by water molecules. At the end of refinement, one glycerol molecule from the cryo-solution was added to both C2 structures. The glycerol molecule in the C2scn structure was found at a corresponding position as in the C2221
scn structure, while the glycerol molecule in the C2urea structure was found in the solvent channel leading towards one of the zinc atoms, where it interacted with surrounding water molecules.
Refinement statistics for the four crystal forms is shown in Table . 95.4% of the residues in the C2221urea structure were found in the most favored regions of the Ramachandran plot and 4.6% in additional allowed regions. For the C2221scn structure, the corresponding values were 95.8% and 4.2% and for the C2 structures (C2urea/C2scn) 96.4%/96.0% and 3.6%/4.0%, respectively. The models showed no residues located in the generously allowed or disallowed regions of the plot.
Data processing and refinement statistics for the orthorhombic and monoclinic insulin crystallized in presence of urea and thiocyanate respectively.
All four insulin molecules are structurally very similar. Pair-wise superposition and comparison of the C2 structures results in a root-mean-square (r.m.s.) distance between corresponding Cα atoms of 0.35 Å and of 0.88 Å when all common atoms are included. For the C2221 structures the same r.m.s. distances are 0.26 Å and 0.58 Å.
A common feature of the four structures is the disruption of the otherwise characteristic continuous a-helix from reside B1 to B19. Instead of having a-helical conformation, some of the PheB1 residues in all four structures have a non-helical conformation. In the C2221 structures, the majority of the B-chains (11/18 and 14/18 in the C2221urea and C2221scn structures respectively) have this conformation (conformation I) where the phi/psi values of ValB2 are -80/+45. In the second conformation (II), the phi/psi values are ~-60/-45, closer to the typical values for an a-helix. The different B-chain conformations are illustrated in Figure , where they are superposed on each other. The distance between the Ca-atom of the PheB1 residue in the two different conformations is ~6 Å. For some of the residues, electron density could be seen for backbone atoms in more than one orientation. In such cases, the conformation with highest density was chosen, where also side chain atoms could be modeled with confidence. It should be noted that the density is weak for the side chain atoms of the Phe1 residue in chains B, F, b, f, h, j and l in the C2221scn structure and chains h, j, l in the C2221urea structure (chain names refer to the continuous naming convention of all chains in the PDB file). The PheB1 orientations in the C2 structures resemble those of the C2221 structures. Three out of the six B-chains in each structure have a non-helical conformation. The electron density is generally better defined in these two structures, which is reflected in the crystallographic B-factors. A comparison of the B-factors shows that PheB1 residues with non-helical conformation have a lower B-factor in three of the four structures, Table .
Figure 2 Superposed B chains from the four structures. (a) C2urea, (b) C2scn, (c) C2221urea and (d) C2221scn. Three B chains in each of the two C2 structures and the majority of the B chains in the C2221 structures have the PheB1 residue in an extended conformation (more ...)
In contrast to the T3
conformation, where B1–B3 have an extended conformation, only resides B1–B2 have a non-helical state. A similar, non-helical conformation of the PheB1 residue has previously been observed for one of the B-chains in an R6
insulin in complex with resorcinol (PDB ID: 1EVR
]. In that case, the carbonyl oxygen of PheB1 is coordinating a sodium ion which was further coordinated by the C terminal AsnA21 of a symmetry-related molecule. In our case, PheB1 is stabilized in a non-helical conformation by a hydrogen bond between amide nitrogen of PheB1 to the carbonyl oxygen of ThrA8 in a neighboring molecule, or a hydrogen bond between the carbonyl oxygen of PheB1 and the amide nitrogen of AsnB3 in the same chain. There are further interactions with symmetry-related molecules, such as the PheB1 amide nitrogen interactions with the OH group of a symmetry-related TyrA14, or the carbonyl oxygen of CysA20 and AsnA18.
In close proximity to the PheB1 residue of the three B chains with non-helical conformation in the C2urea structure there was an electron density peak with a height of 5 σ in the 2Fo-Fc map and ~5 σ in a Fo-Fc difference map. The location of the peak was close to the position where the carbonyl oxygen of PheB1 would be located if the conformation was a-helical. Given the observed electron density, a chloride ion was fitted into this position. It is coordinated to the amide nitrogen of HisB5 (3.2 Å) and two or three water molecules at an average distance of 3.3 Å. The corresponding sites in the C2221urea structure were too disordered to be interpreted in a similar manner.
Location of main-chain and side-chain atoms was ambiguous for the residues LysB29 and ThrB30 in most of the chains in the four structures. Furthermore, the following residues were modeled with alternating side chain conformations; C2urea structure: GlnB4.2, SerB9.5, ValB18.5, LeuB17.6; C2scn structure: GluB13.3, ValB18.4; C2221urea structure: LeuB17.I.1, ArgB22.I.4, GlnB4.I.6, ValB18.II.4 AsnB3.III.1; C2221scn structure: ValB18.I.2, ValB18.II.4, ValB18.II.5 (the roman numerals refer to the hexamer number while the single integer following a punctuation indicates monomer).
There is a strikingly high similarity between the crystallographic contact surfaces of the C2221 and C2 crystal forms. For five of the six contact sites found in the C2 crystal form, there is a corresponding contact surface with equivalent residue composition in the C2221 structure. Each hexamer in the C2221 structure has one symmetry-related contact surface that is identical to the hexamer-hexamer contact in the asymmetric unit. Including the hexamer-hexamer contacts within the asymmetric unit results in five such contact interfaces. In comparison, the C2 structure has in total six neighboring symmetry-related hexamers of which only one has the same kind of pair-wise interactions as the asymmetric hexamer-hexamer contact in the C2221 structure. An overview of the crystal contacts in the C2221 and C2 crystals is shown in Figure .
Overview of the crystal contacts. The crystal contacts in the C2221 (a) and C2 (b) crystals are shown in sphere representation where the interfaces comprising the tyrosine-tyrosine and glutamate-glutamate interaction are shown in red.
A special crystal interaction at the dimer-dimer interface
Each of the two hexamer-hexamer interfaces in the asymmetric unit of the C2221
crystals involves tyrosine-tyrosine interactions between different Tyr A14 groups. Tyr A14 is located at the dimer-dimer interface within the insulin hexamer so that the crystal packing brings four different Tyr A14 groups in proximity, Figure . The tyrosine side chains are pair wise stacked, such that the OH-group of TyrA14 in the first hexamer hydrogen bonds to the backbone oxygen of a TyrA14 in the neighboring hexamer (2.8 Å). The OH-group of the latter TyrA14 forms, in turn, hydrogen bonds to two water molecules. The polar interactions between the hexamers, Figure , comprise hydrogen bonds between GlnA15.I NE2 – GluA17.II OE2 (3.0 Å), GluA17.I OE1 – GlnA15.II NE2 (3.1 Å) (.I or .II denotes different hexamers). Additionally, there is an unusually short contact between two glutamates, GluA17.I OE1 – GluA17.II OE1 (2.32 ± 0.07 Å). In spite of the relative high pH of 6.5, the short Glu-Glu distance suggests a protonated carboxyl group of one of the glutamates. Normally, the pKa value for an exposed glutamate residue is ~4.4 in water environment. Given that GluA17 is protonated, the pKa value must thus be higher. One arginine (ArgB22) is located 2.8 Å from each glutamate and could potentially shift the pKa value by its inductive effect. The pKa
value could also be shifted by the surrounding hydrophobic environment. GluA17 is flanked by the two tyrosine-tyrosine interactions, and it is conceivable that an uncharged protonated glutamate is more favorable in that environment. The short distance is indicative of a strong, low barrier hydrogen bond, where the proton is shared between the two carboxylates. Such low barrier hydrogen bonds have been found in protein active sites as part of enzyme catalysis [22
] but also on protein surfaces [24
Figure 4 One of the hexamer-hexamer interfaces in the C2221 urea structure. The two tyrosine-tyrosine interactions (TyrA14-TyrA14) are flanking a close glutamate-glutamate contact of 2.3 Å. Both glutamates interact with ArgB22 (distance 2.8 Å). (more ...)
As the pH is increased to 7.0, the second crystal form C2 appears. In this crystal form, there is only one crystal packing interaction comprising the tight glutamate-glutamate and tyrosine-tyrosine interaction, Figure . The increased pH could be the reason for the smaller number of such contacts. At higher pH, the shared hydrogen between the two glutamates becomes more delocalized and the repulsive forces will dominate. Consequently, the distance between the carboxylates is longer, 2.40 Å, versus 2.32 Å for the C2221
structures, indicating a weaker interaction at this pH. At pH values above 7.5, only the monoclinic P21
crystal form [11
] is observed, in which no such interface exists.
Interestingly, the position occupied by the tyrosine from a neighboring hexamer is known to bind phenolic compounds like resorcinol and m-cresol [11
]. In Figure the phenolic binding sites in the pdb files 1EVR
hexameric insulin complexed with resorcinol) and 1EV6
hexameric insulin complexed with m-cresol) [11
] are compared with one of the hexamer-hexamer interfaces in the C2221
urea structure. The phenyl ring of the neighboring tyrosine superposes the phenolic derivatives and should contribute to the stability of both the hexamer contact and the insulin structure.
Figure 5 Comparison of the binding pocket for a phenol derivative as seen in other structures and the position for hexamer-hexamer interaction as observed in this study. In (a), the phenolic binding pockets of 1EVR (blue) and 1EV6 (purple) are superposed. One (more ...)
Analysis of crystal contact surfaces
In order to compare the different crystal forms of insulin, the contact sites were characterized by means of polarity and contact area. A summary of the properties for the various contact sites for the four structures presented in this study is shown in Table . Data for other crystal forms of hexameric insulin are also included. The surface area buried by crystal contacts range from 1423 Å 2
to 3314 Å 2
, which constitutes a fraction of buried surface area of between 10.6% and 24%. The smallest value is found for the orthorhombic C2221
crystals where the total contacts surface for the three hexamers is 4269 Å 2
, which amounts to a contact surface of 1423 Å 2
/hexamer. The largest surface area originates from the rhombohedral crystal form, space group R3 with T6
configuration of the B-chain, PDB ID: 1MSO
]. The monoclinic crystals in space group C2 and P21
as well as the tetragonal crystal in space group P43
2 all have six contact sites while the rest have eight. The size of individual contact sites ranges from 236 Å 2
to 414 Å 2
Properties of crystal packing contact surfaces of insulin hexamers in seven different space groups.
The contact surfaces were characterized as either polar (oxygen and nitrogen atoms, including ionisable groups) or non-polar (carbons). The four structures presented in this study constitute a group with a high fraction of non-polar contact surface, ranging from 53% to 56% of the total contact area, compared to 41% to 50% for the other crystal forms. The monoclinic P21 crystal form is the most hydrophilic, with a 40/60 distribution between hydrophobic and hydrophilic contact area. This analysis is limited in that bound water molecules were not considered in the crystal contact interactions since the criteria for modeling water molecules may vary among crystallographers and are also dependent on data quality. Several interactions could however involve hydrogen bonds to water molecules. Side chains with missing atoms were rebuilt in order to use the surface with an atom composition representing the true surface for the property calculations. They were however rebuilt automatically and could potentially be in a wrong orientation.
Comparing the residue identity of the crystal contacts for the crystals presented in Table shows that seven of the interface residues are common for all crystal forms (GlnA5, ThrA8, TyrA14, GlnA15, AsnA18, TyrA19 and PheB1). Altogether, the contact sites for the six crystal types compared in this study cover almost the entire surface of an insulin hexamer. A comparison of the exposed residues with the residues involved in crystal contacts shows that all residues with an exposure of more than 20% participate in some contact interface. The degree of exposure was calculated according to [25
]. A number of studies, where crystal packing contacts have been systematically investigated [26
] conclude that atomic composition within crystal contacts is indistinguishable from that of the protein surface and is rather non-specific. Studies of pancreatic ribonuclease [28
] and cutinase [29
], crystallized in a number of space groups, showed in accordance with the present study that virtually the entirely protein surface can be involved in crystal contacts.
The C2221urea and C2urea crystals were grown in presence of 3 M and 4 M urea, respectively. Seven urea molecules were built into the C2 structure. Five of these were located at equivalent positions in the monomers, Figure . The nitrogen atoms hydrogen bond primarily to the carbonyl oxygen of GlnA5, but the carbonyl oxygen of SerA9 and IleA10 are also within a reasonable hydrogen bonding distance (average 3.1 Å). In monomer six, the urea is either disordered or not present. Instead, a water molecule was built into the density. Nine out of 18 possible equivalent positions in the C2221 structure were occupied by urea. In the nine positions without a urea molecule, water was built in. Furthermore, these positions are more distant to a neighboring hexamer and therefore have a less well-defined electron density which may explain the inability to model a urea molecule. Exceptions from the above generalization are the monomers II.4 and III.1, which are close to a neighboring hexamer, but the electron density indicates two ordered water molecules.
Figure 6 Urea binding site. The most commonly occupied binding site for the urea molecule in the C2urea and C2221urea structures. Hydrogen bonds are primarily directed towards the carbonyl oxygen GlnA5 but surrounding carbonyl oxygens from SerA9 and IleA10 are (more ...)
Insulin has been shown to be tolerant of high concentrations of urea and other denaturants [30
] and urea has previously been used to increase its solubility. One example is the inclusion of urea to promote growth of larger crystals of an insulin-protamine complex [32
]. No significant conformational changes were detected in our structures as a result of direct urea interactions. In spite of its common use as a chemical denaturant of proteins, the molecular mechanism of urea-mediated unfolding is not known. Identification of denaturant interactions with proteins may give insight into the early stages of protein unfolding [33
]. For a denaturant to be effective, protein-solvent interactions must be disturbed and this is thought to happen either through a direct or an indirect mechanism [34
]. A direct mechanism would involve binding of urea molecules to the protein surface and thus compete with water-protein interactions and enhance the solubility of hydrophobic residues. Indirect urea denaturation would involve disruption of solvent-mediated hydrophobic interactions which would destabilize the protein structure. In addition, studies have shown that urea and guanidine hydrochloride at sub-denaturant concentrations stabilizes proteins at a sub global level in a mechanism called protein stiffening [35
]. The present study shows that urea at concentrations ~3 M has one specific binding site on the surface of the insulin molecule, interacting with backbone carbonyl groups of primarily GlnA5 but also of SerA9 and IleA10 residues. Given the high concentrations of urea present in the crystallization experiments, we would expect to detect even weak binding sites with a Kd
of several hundred mM. Thus it seems unlikely that insulin denaturation occurs via a direct mechanism which requires binding of several urea molecules. On the other hand we see no signs of partial unfolding in our structures which would be indicative of an indirect mechanism. A recent study suggests that the denaturant effect of urea is neither due to a direct or indirect mechanism but rather an effect of a reduction of ion pairing between ionic and polar groups at aggregate surfaces [37
], something which also could explain the relative higher fraction of non-polar surface at crystal contacts in our crystal forms.